This document describes filters, sources, and sinks provided by the libavfilter library.
Filtering in FFmpeg is enabled through the libavfilter library.
In libavfilter, a filter can have multiple inputs and multiple outputs. To illustrate the sorts of things that are possible, we consider the following filtergraph.
[main] input --> split ---------------------> overlay --> output | ^ |[tmp] [flip]| +-----> crop --> vflip -------+
This filtergraph splits the input stream in two streams, then sends one stream through the crop filter and the vflip filter, before merging it back with the other stream by overlaying it on top. You can use the following command to achieve this:
ffmpeg -i INPUT -vf "split [main][tmp]; [tmp] crop=iw:ih/2:0:0, vflip [flip]; [main][flip] overlay=0:H/2" OUTPUT
The result will be that the top half of the video is mirrored onto the bottom half of the output video.
Filters in the same linear chain are separated by commas, and distinct linear chains of filters are separated by semicolons. In our example, crop,vflip are in one linear chain, split and overlay are separately in another. The points where the linear chains join are labelled by names enclosed in square brackets. In the example, the split filter generates two outputs that are associated to the labels [main] and [tmp].
The stream sent to the second output of split, labelled as [tmp], is processed through the crop filter, which crops away the lower half part of the video, and then vertically flipped. The overlay filter takes in input the first unchanged output of the split filter (which was labelled as [main]), and overlay on its lower half the output generated by the crop,vflip filterchain.
Some filters take in input a list of parameters: they are specified after the filter name and an equal sign, and are separated from each other by a colon.
There exist so-called source filters that do not have an audio/video input, and sink filters that will not have audio/video output.
The graph2dot program included in the FFmpeg tools directory can be used to parse a filtergraph description and issue a corresponding textual representation in the dot language.
Invoke the command:
graph2dot -h
to see how to use graph2dot.
You can then pass the dot description to the dot program (from the graphviz suite of programs) and obtain a graphical representation of the filtergraph.
For example the sequence of commands:
echo GRAPH_DESCRIPTION | \ tools/graph2dot -o graph.tmp && \ dot -Tpng graph.tmp -o graph.png && \ display graph.png
can be used to create and display an image representing the graph described by the GRAPH_DESCRIPTION string. Note that this string must be a complete self-contained graph, with its inputs and outputs explicitly defined. For example if your command line is of the form:
ffmpeg -i infile -vf scale=640:360 outfile
your GRAPH_DESCRIPTION string will need to be of the form:
nullsrc,scale=640:360,nullsink
you may also need to set the nullsrc parameters and add a format filter in order to simulate a specific input file.
A filtergraph is a directed graph of connected filters. It can contain cycles, and there can be multiple links between a pair of filters. Each link has one input pad on one side connecting it to one filter from which it takes its input, and one output pad on the other side connecting it to one filter accepting its output.
Each filter in a filtergraph is an instance of a filter class registered in the application, which defines the features and the number of input and output pads of the filter.
A filter with no input pads is called a "source", and a filter with no output pads is called a "sink".
A filtergraph has a textual representation, which is recognized by the
-filter/-vf/-af and
-filter_complex options in ffmpeg
and
-vf/-af in ffplay
, and by the
avfilter_graph_parse_ptr()
function defined in
libavfilter/avfilter.h.
A filterchain consists of a sequence of connected filters, each one connected to the previous one in the sequence. A filterchain is represented by a list of ","-separated filter descriptions.
A filtergraph consists of a sequence of filterchains. A sequence of filterchains is represented by a list of ";"-separated filterchain descriptions.
A filter is represented by a string of the form: [in_link_1]...[in_link_N]filter_name@id=arguments[out_link_1]...[out_link_M]
filter_name is the name of the filter class of which the described filter is an instance of, and has to be the name of one of the filter classes registered in the program optionally followed by "@id". The name of the filter class is optionally followed by a string "=arguments".
arguments is a string which contains the parameters used to initialize the filter instance. It may have one of two forms:
fade
filter
declares three options in this order – type, start_frame and
nb_frames. Then the parameter list in:0:30 means that the value
in is assigned to the option type, 0 to
start_frame and 30 to nb_frames.
If the option value itself is a list of items (e.g. the format
filter
takes a list of pixel formats), the items in the list are usually separated by
‘|’.
The list of arguments can be quoted using the character ‘'’ as initial and ending mark, and the character ‘\’ for escaping the characters within the quoted text; otherwise the argument string is considered terminated when the next special character (belonging to the set ‘[]=;,’) is encountered.
A special syntax implemented in the ffmpeg
CLI tool allows loading
option values from files. This is done be prepending a slash ’/’ to the option
name, then the supplied value is interpreted as a path from which the actual
value is loaded. E.g.
ffmpeg -i <INPUT> -vf drawtext=/text=/tmp/some_text <OUTPUT>
will load the text to be drawn from /tmp/some_text. API users wishing to
implement a similar feature should use the avfilter_graph_segment_*()
functions together with custom IO code.
The name and arguments of the filter are optionally preceded and followed by a list of link labels. A link label allows one to name a link and associate it to a filter output or input pad. The preceding labels in_link_1 ... in_link_N, are associated to the filter input pads, the following labels out_link_1 ... out_link_M, are associated to the output pads.
When two link labels with the same name are found in the filtergraph, a link between the corresponding input and output pad is created.
If an output pad is not labelled, it is linked by default to the first unlabelled input pad of the next filter in the filterchain. For example in the filterchain
nullsrc, split[L1], [L2]overlay, nullsink
the split filter instance has two output pads, and the overlay filter instance two input pads. The first output pad of split is labelled "L1", the first input pad of overlay is labelled "L2", and the second output pad of split is linked to the second input pad of overlay, which are both unlabelled.
In a filter description, if the input label of the first filter is not specified, "in" is assumed; if the output label of the last filter is not specified, "out" is assumed.
In a complete filterchain all the unlabelled filter input and output pads must be connected. A filtergraph is considered valid if all the filter input and output pads of all the filterchains are connected.
Leading and trailing whitespaces (space, tabs, or line feeds) separating tokens in the filtergraph specification are ignored. This means that the filtergraph can be expressed using empty lines and spaces to improve redability.
For example, the filtergraph:
testsrc,split[L1],hflip[L2];[L1][L2] hstack
can be represented as:
testsrc, split [L1], hflip [L2]; [L1][L2] hstack
Libavfilter will automatically insert scale filters where format
conversion is required. It is possible to specify swscale flags
for those automatically inserted scalers by prepending
sws_flags=flags;
to the filtergraph description.
Here is a BNF description of the filtergraph syntax:
NAME ::= sequence of alphanumeric characters and '_' FILTER_NAME ::= NAME["@"NAME] LINKLABEL ::= "[" NAME "]" LINKLABELS ::= LINKLABEL [LINKLABELS] FILTER_ARGUMENTS ::= sequence of chars (possibly quoted) FILTER ::= [LINKLABELS] FILTER_NAME ["=" FILTER_ARGUMENTS] [LINKLABELS] FILTERCHAIN ::= FILTER [,FILTERCHAIN] FILTERGRAPH ::= [sws_flags=flags;] FILTERCHAIN [;FILTERGRAPH]
Filtergraph description composition entails several levels of escaping. See (ffmpeg-utils)the "Quoting and escaping" section in the ffmpeg-utils(1) manual for more information about the employed escaping procedure.
A first level escaping affects the content of each filter option
value, which may contain the special character :
used to
separate values, or one of the escaping characters \'
.
A second level escaping affects the whole filter description, which
may contain the escaping characters \'
or the special
characters [],;
used by the filtergraph description.
Finally, when you specify a filtergraph on a shell commandline, you need to perform a third level escaping for the shell special characters contained within it.
For example, consider the following string to be embedded in the drawtext filter description text value:
this is a 'string': may contain one, or more, special characters
This string contains the '
special escaping character, and the
:
special character, so it needs to be escaped in this way:
text=this is a \'string\'\: may contain one, or more, special characters
A second level of escaping is required when embedding the filter description in a filtergraph description, in order to escape all the filtergraph special characters. Thus the example above becomes:
drawtext=text=this is a \\\'string\\\'\\: may contain one\, or more\, special characters
(note that in addition to the \'
escaping special characters,
also ,
needs to be escaped).
Finally an additional level of escaping is needed when writing the
filtergraph description in a shell command, which depends on the
escaping rules of the adopted shell. For example, assuming that
\
is special and needs to be escaped with another \
, the
previous string will finally result in:
-vf "drawtext=text=this is a \\\\\\'string\\\\\\'\\\\: may contain one\\, or more\\, special characters"
In order to avoid cumbersome escaping when using a commandline tool accepting a filter specification as input, it is advisable to avoid direct inclusion of the filter or options specification in the shell.
For example, in case of the drawtext filter, you might prefer to use the textfile option in place of text to specify the text to render.
When using the ffmpeg
tool, you might consider to use the
(ffmpeg)-filter_script option or
(ffmpeg)-filter_complex_script option.
Some filters support a generic enable option. For the filters supporting timeline editing, this option can be set to an expression which is evaluated before sending a frame to the filter. If the evaluation is non-zero, the filter will be enabled, otherwise the frame will be sent unchanged to the next filter in the filtergraph.
The expression accepts the following values:
timestamp expressed in seconds, NAN if the input timestamp is unknown
sequential number of the input frame, starting from 0
the position in the file of the input frame, NAN if unknown; deprecated, do not use
width and height of the input frame if video
Additionally, these filters support an enable command that can be used to re-define the expression.
Like any other filtering option, the enable option follows the same rules.
For example, to enable a blur filter (smartblur) from 10 seconds to 3 minutes, and a curves filter starting at 3 seconds:
smartblur = enable='between(t,10,3*60)', curves = enable='gte(t,3)' : preset=cross_process
See ffmpeg -filters
to view which filters have timeline support.
Some options can be changed during the operation of the filter using
a command. These options are marked ’T’ on the output of
ffmpeg
-h filter=<name of filter>.
The name of the command is the name of the option and the argument is
the new value.
Some filters with several inputs support a common set of options. These options can only be set by name, not with the short notation.
The action to take when EOF is encountered on the secondary input; it accepts one of the following values:
Repeat the last frame (the default).
End both streams.
Pass the main input through.
If set to 1, force the output to terminate when the shortest input terminates. Default value is 0.
If set to 1, force the filter to extend the last frame of secondary streams until the end of the primary stream. A value of 0 disables this behavior. Default value is 1.
How strictly to sync streams based on secondary input timestamps; it accepts one of the following values:
Frame from secondary input with the nearest lower or equal timestamp to the primary input frame.
Frame from secondary input with the absolute nearest timestamp to the primary input frame.
When you configure your FFmpeg build, you can disable any of the
existing filters using --disable-filters
.
The configure output will show the audio filters included in your
build.
Below is a description of the currently available audio filters.
Apply Affine Projection algorithm to the first audio stream using the second audio stream.
This adaptive filter is used to estimate unknown audio based on multiple input audio samples. Affine projection algorithm can make trade-offs between computation complexity with convergence speed.
A description of the accepted options follows.
Set the filter order.
Set the projection order.
Set the filter mu.
Set the coefficient to initialize internal covariance matrix.
Set the filter output samples. It accepts the following values:
Pass the 1st input.
Pass the 2nd input.
Pass difference between desired, 2nd input and error signal estimate.
Pass difference between input, 1st input and error signal estimate.
Pass error signal estimated samples.
Default value is o.
Set which precision to use when processing samples.
Auto pick internal sample format depending on other filters.
Always use single-floating point precision sample format.
Always use double-floating point precision sample format.
A compressor is mainly used to reduce the dynamic range of a signal. Especially modern music is mostly compressed at a high ratio to improve the overall loudness. It’s done to get the highest attention of a listener, "fatten" the sound and bring more "power" to the track. If a signal is compressed too much it may sound dull or "dead" afterwards or it may start to "pump" (which could be a powerful effect but can also destroy a track completely). The right compression is the key to reach a professional sound and is the high art of mixing and mastering. Because of its complex settings it may take a long time to get the right feeling for this kind of effect.
Compression is done by detecting the volume above a chosen level
threshold
and dividing it by the factor set with ratio
.
So if you set the threshold to -12dB and your signal reaches -6dB a ratio
of 2:1 will result in a signal at -9dB. Because an exact manipulation of
the signal would cause distortion of the waveform the reduction can be
levelled over the time. This is done by setting "Attack" and "Release".
attack
determines how long the signal has to rise above the threshold
before any reduction will occur and release
sets the time the signal
has to fall below the threshold to reduce the reduction again. Shorter signals
than the chosen attack time will be left untouched.
The overall reduction of the signal can be made up afterwards with the
makeup
setting. So compressing the peaks of a signal about 6dB and
raising the makeup to this level results in a signal twice as loud than the
source. To gain a softer entry in the compression the knee
flattens the
hard edge at the threshold in the range of the chosen decibels.
The filter accepts the following options:
Set input gain. Default is 1. Range is between 0.015625 and 64.
Set mode of compressor operation. Can be upward
or downward
.
Default is downward
.
If a signal of stream rises above this level it will affect the gain reduction. By default it is 0.125. Range is between 0.00097563 and 1.
Set a ratio by which the signal is reduced. 1:2 means that if the level rose 4dB above the threshold, it will be only 2dB above after the reduction. Default is 2. Range is between 1 and 20.
Amount of milliseconds the signal has to rise above the threshold before gain reduction starts. Default is 20. Range is between 0.01 and 2000.
Amount of milliseconds the signal has to fall below the threshold before reduction is decreased again. Default is 250. Range is between 0.01 and 9000.
Set the amount by how much signal will be amplified after processing. Default is 1. Range is from 1 to 64.
Curve the sharp knee around the threshold to enter gain reduction more softly. Default is 2.82843. Range is between 1 and 8.
Choose if the average
level between all channels of input stream
or the louder(maximum
) channel of input stream affects the
reduction. Default is average
.
Should the exact signal be taken in case of peak
or an RMS one in case
of rms
. Default is rms
which is mostly smoother.
How much to use compressed signal in output. Default is 1. Range is between 0 and 1.
This filter supports the all above options as commands.
Simple audio dynamic range compression/expansion filter.
The filter accepts the following options:
Set contrast. Default is 33. Allowed range is between 0 and 100.
Copy the input audio source unchanged to the output. This is mainly useful for testing purposes.
Apply cross fade from one input audio stream to another input audio stream. The cross fade is applied for specified duration near the end of first stream.
The filter accepts the following options:
Specify the number of samples for which the cross fade effect has to last. At the end of the cross fade effect the first input audio will be completely silent. Default is 44100.
Specify the duration of the cross fade effect. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax. By default the duration is determined by nb_samples. If set this option is used instead of nb_samples.
Should first stream end overlap with second stream start. Default is enabled.
Set curve for cross fade transition for first stream.
Set curve for cross fade transition for second stream.
For description of available curve types see afade filter description.
ffmpeg -i first.flac -i second.flac -filter_complex acrossfade=d=10:c1=exp:c2=exp output.flac
ffmpeg -i first.flac -i second.flac -filter_complex acrossfade=d=10:o=0:c1=exp:c2=exp output.flac
Split audio stream into several bands.
This filter splits audio stream into two or more frequency ranges. Summing all streams back will give flat output.
The filter accepts the following options:
Set split frequencies. Those must be positive and increasing.
Set filter order for each band split. This controls filter roll-off or steepness of filter transfer function. Available values are:
12 dB per octave.
24 dB per octave.
36 dB per octave.
48 dB per octave.
60 dB per octave.
72 dB per octave.
84 dB per octave.
96 dB per octave.
108 dB per octave.
120 dB per octave.
Default is 4th.
Set input gain level. Allowed range is from 0 to 1. Default value is 1.
Set output gain for each band. Default value is 1 for all bands.
Set which precision to use when processing samples.
Auto pick internal sample format depending on other filters.
Always use single-floating point precision sample format.
Always use double-floating point precision sample format.
Default value is auto
.
ffmpeg -i in.flac -filter_complex 'acrossover=split=1500[LOW][HIGH]' -map '[LOW]' low.wav -map '[HIGH]' high.wav
ffmpeg -i in.flac -filter_complex 'acrossover=split=1500:order=8th[LOW][HIGH]' -map '[LOW]' low.wav -map '[HIGH]' high.wav
ffmpeg -i in.flac -filter_complex 'acrossover=split=1500 8000:order=8th[LOW][MID][HIGH]' -map '[LOW]' low.wav -map '[MID]' mid.wav -map '[HIGH]' high.wav
Reduce audio bit resolution.
This filter is bit crusher with enhanced functionality. A bit crusher is used to audibly reduce number of bits an audio signal is sampled with. This doesn’t change the bit depth at all, it just produces the effect. Material reduced in bit depth sounds more harsh and "digital". This filter is able to even round to continuous values instead of discrete bit depths. Additionally it has a D/C offset which results in different crushing of the lower and the upper half of the signal. An Anti-Aliasing setting is able to produce "softer" crushing sounds.
Another feature of this filter is the logarithmic mode. This setting switches from linear distances between bits to logarithmic ones. The result is a much more "natural" sounding crusher which doesn’t gate low signals for example. The human ear has a logarithmic perception, so this kind of crushing is much more pleasant. Logarithmic crushing is also able to get anti-aliased.
The filter accepts the following options:
Set level in.
Set level out.
Set bit reduction.
Set mixing amount.
Can be linear: lin
or logarithmic: log
.
Set DC.
Set anti-aliasing.
Set sample reduction.
Enable LFO. By default disabled.
Set LFO range.
Set LFO rate.
This filter supports the all above options as commands.
Delay audio filtering until a given wallclock timestamp. See the cue filter.
Remove impulsive noise from input audio.
Samples detected as impulsive noise are replaced by interpolated samples using autoregressive modelling.
Set window size, in milliseconds. Allowed range is from 10
to
100
. Default value is 55
milliseconds.
This sets size of window which will be processed at once.
Set window overlap, in percentage of window size. Allowed range is from
50
to 95
. Default value is 75
percent.
Setting this to a very high value increases impulsive noise removal but makes
whole process much slower.
Set autoregression order, in percentage of window size. Allowed range is from
0
to 25
. Default value is 2
percent. This option also
controls quality of interpolated samples using neighbour good samples.
Set threshold value. Allowed range is from 1
to 100
.
Default value is 2
.
This controls the strength of impulsive noise which is going to be removed.
The lower value, the more samples will be detected as impulsive noise.
Set burst fusion, in percentage of window size. Allowed range is 0
to
10
. Default value is 2
.
If any two samples detected as noise are spaced less than this value then any
sample between those two samples will be also detected as noise.
Set overlap method.
It accepts the following values:
Select overlap-add method. Even not interpolated samples are slightly changed with this method.
Select overlap-save method. Not interpolated samples remain unchanged.
Default value is a
.
Remove clipped samples from input audio.
Samples detected as clipped are replaced by interpolated samples using autoregressive modelling.
Set window size, in milliseconds. Allowed range is from 10
to 100
.
Default value is 55
milliseconds.
This sets size of window which will be processed at once.
Set window overlap, in percentage of window size. Allowed range is from 50
to 95
. Default value is 75
percent.
Set autoregression order, in percentage of window size. Allowed range is from
0
to 25
. Default value is 8
percent. This option also controls
quality of interpolated samples using neighbour good samples.
Set threshold value. Allowed range is from 1
to 100
.
Default value is 10
. Higher values make clip detection less aggressive.
Set size of histogram used to detect clips. Allowed range is from 100
to 9999
.
Default value is 1000
. Higher values make clip detection less aggressive.
Set overlap method.
It accepts the following values:
Select overlap-add method. Even not interpolated samples are slightly changed with this method.
Select overlap-save method. Not interpolated samples remain unchanged.
Default value is a
.
Apply decorrelation to input audio stream.
The filter accepts the following options:
Set decorrelation stages of filtering. Allowed range is from 1 to 16. Default value is 6.
Set random seed used for setting delay in samples across channels.
Delay one or more audio channels.
Samples in delayed channel are filled with silence.
The filter accepts the following option:
Set list of delays in milliseconds for each channel separated by ’|’. Unused delays will be silently ignored. If number of given delays is smaller than number of channels all remaining channels will not be delayed. If you want to delay exact number of samples, append ’S’ to number. If you want instead to delay in seconds, append ’s’ to number.
Use last set delay for all remaining channels. By default is disabled.
This option if enabled changes how option delays
is interpreted.
adelay=1500|0|500
adelay=0|500S|700S
adelay=delays=64S:all=1
Remedy denormals in audio by adding extremely low-level noise.
This filter shall be placed before any filter that can produce denormals.
A description of the accepted parameters follows.
Set level of added noise in dB. Default is -351
.
Allowed range is from -451 to -90.
Set type of added noise.
Add DC signal.
Add AC signal.
Add square signal.
Add pulse signal.
Default is dc
.
This filter supports the all above options as commands.
Compute derivative/integral of audio stream.
Applying both filters one after another produces original audio.
Apply spectral dynamic range controller filter to input audio stream.
A description of the accepted options follows.
Set the transfer expression.
The expression can contain the following constants:
current channel number
current sample number
number of channels
timestamp expressed in seconds
sample rate
current frequency power value, in dB
current frequency in Hz
Default value is p
.
Set the attack in milliseconds. Default is 50
milliseconds.
Allowed range is from 1 to 1000 milliseconds.
Set the release in milliseconds. Default is 100
milliseconds.
Allowed range is from 5 to 2000 milliseconds.
Set which channels to filter, by default all
channels in audio stream are filtered.
This filter supports the all above options as commands.
adrc=transfer='if(gt(p,-50),-50+(p-(-50))/6,p)':attack=50:release=100
adrc=transfer='if(gt(p,-50),-50+(p-(-50))/2,p)':attack=50:release=100:channels=FC
adrc=transfer='if(lte(p,-85),p-800,p)':attack=1:release=5
adrc=transfer='if(lt(p,-10),-10+(p-(-10))*2,p)':attack=50:release=100
adrc=transfer='min(p,-60)':attack=2:release=10
Apply dynamic equalization to input audio stream.
A description of the accepted options follows.
Set the detection threshold used to trigger equalization. Threshold detection is using detection filter. Default value is 0. Allowed range is from 0 to 100.
Set the detection frequency in Hz used for detection filter used to trigger equalization. Default value is 1000 Hz. Allowed range is between 2 and 1000000 Hz.
Set the detection resonance factor for detection filter used to trigger equalization. Default value is 1. Allowed range is from 0.001 to 1000.
Set the target frequency of equalization filter. Default value is 1000 Hz. Allowed range is between 2 and 1000000 Hz.
Set the target resonance factor for target equalization filter. Default value is 1. Allowed range is from 0.001 to 1000.
Set the amount of milliseconds the signal from detection has to rise above the detection threshold before equalization starts. Default is 20. Allowed range is between 1 and 2000.
Set the amount of milliseconds the signal from detection has to fall below the detection threshold before equalization ends. Default is 200. Allowed range is between 1 and 2000.
Set the ratio by which the equalization gain is raised. Default is 1. Allowed range is between 0 and 30.
Set the makeup offset by which the equalization gain is raised. Default is 0. Allowed range is between 0 and 100.
Set the max allowed cut/boost amount. Default is 50. Allowed range is from 1 to 200.
Set the mode of filter operation, can be one of the following:
Output only isolated detection signal.
Cut frequencies below detection threshold.
Cut frequencies above detection threshold.
Boost frequencies below detection threshold.
Boost frequencies above detection threshold.
Default mode is ‘cutbelow’.
Set the type of detection filter, can be one of the following:
Default type is ‘bandpass’.
Set the type of target filter, can be one of the following:
Default type is ‘bell’.
Automatically gather threshold from detection filter. By default is ‘disabled’. This option is useful to detect threshold in certain time frame of input audio stream, in such case option value is changed at runtime.
Available values are:
Disable using automatically gathered threshold value.
Stop picking threshold value.
Start picking threshold value.
Adaptively pick threshold value, by calculating sliding window entropy.
Set which precision to use when processing samples.
Auto pick internal sample format depending on other filters.
Always use single-floating point precision sample format.
Always use double-floating point precision sample format.
This filter supports the all above options as commands.
Apply dynamic smoothing to input audio stream.
A description of the accepted options follows.
Set an amount of sensitivity to frequency fluctations. Default is 2. Allowed range is from 0 to 1e+06.
Set a base frequency for smoothing. Default value is 22050. Allowed range is from 2 to 1e+06.
This filter supports the all above options as commands.
Apply echoing to the input audio.
Echoes are reflected sound and can occur naturally amongst mountains
(and sometimes large buildings) when talking or shouting; digital echo
effects emulate this behaviour and are often used to help fill out the
sound of a single instrument or vocal. The time difference between the
original signal and the reflection is the delay
, and the
loudness of the reflected signal is the decay
.
Multiple echoes can have different delays and decays.
A description of the accepted parameters follows.
Set input gain of reflected signal. Default is 0.6
.
Set output gain of reflected signal. Default is 0.3
.
Set list of time intervals in milliseconds between original signal and reflections
separated by ’|’. Allowed range for each delay
is (0 - 90000.0]
.
Default is 1000
.
Set list of loudness of reflected signals separated by ’|’.
Allowed range for each decay
is (0 - 1.0]
.
Default is 0.5
.
aecho=0.8:0.88:60:0.4
aecho=0.8:0.88:6:0.4
aecho=0.8:0.9:1000:0.3
aecho=0.8:0.9:1000|1800:0.3|0.25
Audio emphasis filter creates or restores material directly taken from LPs or emphased CDs with different filter curves. E.g. to store music on vinyl the signal has to be altered by a filter first to even out the disadvantages of this recording medium. Once the material is played back the inverse filter has to be applied to restore the distortion of the frequency response.
The filter accepts the following options:
Set input gain.
Set output gain.
Set filter mode. For restoring material use reproduction
mode, otherwise
use production
mode. Default is reproduction
mode.
Set filter type. Selects medium. Can be one of the following:
select Columbia.
select EMI.
select BSI (78RPM).
select RIAA.
select Compact Disc (CD).
select 50µs (FM).
select 75µs (FM).
select 50µs (FM-KF).
select 75µs (FM-KF).
This filter supports the all above options as commands.
Modify an audio signal according to the specified expressions.
This filter accepts one or more expressions (one for each channel), which are evaluated and used to modify a corresponding audio signal.
It accepts the following parameters:
Set the ’|’-separated expressions list for each separate channel. If the number of input channels is greater than the number of expressions, the last specified expression is used for the remaining output channels.
Set output channel layout. If not specified, the channel layout is specified by the number of expressions. If set to ‘same’, it will use by default the same input channel layout.
Each expression in exprs can contain the following constants and functions:
channel number of the current expression
number of the evaluated sample, starting from 0
sample rate
time of the evaluated sample expressed in seconds
input and output number of channels
the value of input channel with number CH
Note: this filter is slow. For faster processing you should use a dedicated filter.
aeval=val(ch)/2:c=same
aeval=val(0)|-val(1)
An exciter is used to produce high sound that is not present in the original signal. This is done by creating harmonic distortions of the signal which are restricted in range and added to the original signal. An Exciter raises the upper end of an audio signal without simply raising the higher frequencies like an equalizer would do to create a more "crisp" or "brilliant" sound.
The filter accepts the following options:
Set input level prior processing of signal. Allowed range is from 0 to 64. Default value is 1.
Set output level after processing of signal. Allowed range is from 0 to 64. Default value is 1.
Set the amount of harmonics added to original signal. Allowed range is from 0 to 64. Default value is 1.
Set the amount of newly created harmonics. Allowed range is from 0.1 to 10. Default value is 8.5.
Set the octave of newly created harmonics. Allowed range is from -10 to 10. Default value is 0.
Set the lower frequency limit of producing harmonics in Hz. Allowed range is from 2000 to 12000 Hz. Default is 7500 Hz.
Set the upper frequency limit of producing harmonics. Allowed range is from 9999 to 20000 Hz. If value is lower than 10000 Hz no limit is applied.
Mute the original signal and output only added harmonics. By default is disabled.
Apply fade-in/out effect to input audio.
A description of the accepted parameters follows.
Specify the effect type, can be either in
for fade-in, or
out
for a fade-out effect. Default is in
.
Specify the number of the start sample for starting to apply the fade effect. Default is 0.
Specify the number of samples for which the fade effect has to last. At the end of the fade-in effect the output audio will have the same volume as the input audio, at the end of the fade-out transition the output audio will be silence. Default is 44100.
Specify the start time of the fade effect. Default is 0. The value must be specified as a time duration; see (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax. If set this option is used instead of start_sample.
Specify the duration of the fade effect. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax. At the end of the fade-in effect the output audio will have the same volume as the input audio, at the end of the fade-out transition the output audio will be silence. By default the duration is determined by nb_samples. If set this option is used instead of nb_samples.
Set curve for fade transition.
It accepts the following values:
select triangular, linear slope (default)
select quarter of sine wave
select half of sine wave
select exponential sine wave
select logarithmic
select inverted parabola
select quadratic
select cubic
select square root
select cubic root
select parabola
select exponential
select inverted quarter of sine wave
select inverted half of sine wave
select double-exponential seat
select double-exponential sigmoid
select logistic sigmoid
select sine cardinal function
select inverted sine cardinal function
select quartic
select quartic root
select squared quarter of sine wave
select squared half of sine wave
no fade applied
Set the initial gain for fade-in or final gain for fade-out.
Default value is 0.0
.
Set the initial gain for fade-out or final gain for fade-in.
Default value is 1.0
.
This filter supports the all above options as commands.
afade=t=in:ss=0:d=15
afade=t=out:st=875:d=25
Denoise audio samples with FFT.
A description of the accepted parameters follows.
Set the noise reduction in dB, allowed range is 0.01 to 97. Default value is 12 dB.
Set the noise floor in dB, allowed range is -80 to -20. Default value is -50 dB.
Set the noise type.
It accepts the following values:
Select white noise.
Select vinyl noise.
Select shellac noise.
Select custom noise, defined in bn
option.
Default value is white noise.
Set custom band noise profile for every one of 15 bands. Bands are separated by ’ ’ or ’|’.
Set the residual floor in dB, allowed range is -80 to -20. Default value is -38 dB.
Enable noise floor tracking. By default is disabled. With this enabled, noise floor is automatically adjusted.
Enable residual tracking. By default is disabled.
Set the output mode.
It accepts the following values:
Pass input unchanged.
Pass noise filtered out.
Pass only noise.
Default value is output.
Set the adaptivity factor, used how fast to adapt gains adjustments per each frequency bin. Value 0 enables instant adaptation, while higher values react much slower. Allowed range is from 0 to 1. Default value is 0.5.
Set the noise floor offset factor. This option is used to adjust offset applied to measured noise floor. It is only effective when noise floor tracking is enabled. Allowed range is from -2.0 to 2.0. Default value is 1.0.
Set the noise link used for multichannel audio.
It accepts the following values:
Use unchanged channel’s noise floor.
Use measured min noise floor of all channels.
Use measured max noise floor of all channels.
Use measured average noise floor of all channels.
Default value is min.
Set the band multiplier factor, used how much to spread bands across frequency bins. Allowed range is from 0.2 to 5. Default value is 1.25.
Toggle capturing and measurement of noise profile from input audio.
It accepts the following values:
Start sample noise capture.
Stop sample noise capture and measure new noise band profile.
Default value is none
.
Set gain smooth spatial radius, used to smooth gains applied to each frequency bin.
Useful to reduce random music noise artefacts.
Higher values increases smoothing of gains.
Allowed range is from 0
to 50
.
Default value is 0
.
This filter supports the some above mentioned options as commands.
afftdn=nr=10:nf=-40
afftdn=nr=10:nf=-80:tn=1
asendcmd=0.0 afftdn sn start,asendcmd=0.4 afftdn sn stop,afftdn=nr=20:nf=-40
Apply arbitrary expressions to samples in frequency domain.
Set frequency domain real expression for each separate channel separated by ’|’. Default is "re". If the number of input channels is greater than the number of expressions, the last specified expression is used for the remaining output channels.
Set frequency domain imaginary expression for each separate channel separated by ’|’. Default is "im".
Each expression in real and imag can contain the following constants and functions:
sample rate
current frequency bin number
number of available bins
channel number of the current expression
number of channels
current frame pts
current real part of frequency bin of current channel
current imaginary part of frequency bin of current channel
Return the value of real part of frequency bin at location (bin,channel)
Return the value of imaginary part of frequency bin at location (bin,channel)
Set window size. Allowed range is from 16 to 131072.
Default is 4096
Set window function.
It accepts the following values:
Default is hann
.
Set window overlap. If set to 1, the recommended overlap for selected
window function will be picked. Default is 0.75
.
afftfilt="'real=re * (1-clip((b/nb)*b,0,1))':imag='im * (1-clip((b/nb)*b,0,1))'"
afftfilt="real='hypot(re,im)*sin(0)':imag='hypot(re,im)*cos(0)':win_size=512:overlap=0.75"
afftfilt="real='hypot(re,im)*cos((random(0)*2-1)*2*3.14)':imag='hypot(re,im)*sin((random(1)*2-1)*2*3.14)':win_size=128:overlap=0.8"
afftfilt="real=re*cos(1)-im*sin(1):imag=re*sin(1)+im*cos(1)"
Apply an arbitrary Finite Impulse Response filter.
This filter is designed for applying long FIR filters, up to 60 seconds long.
It can be used as component for digital crossover filters, room equalization, cross talk cancellation, wavefield synthesis, auralization, ambiophonics, ambisonics and spatialization.
This filter uses the streams higher than first one as FIR coefficients. If the non-first stream holds a single channel, it will be used for all input channels in the first stream, otherwise the number of channels in the non-first stream must be same as the number of channels in the first stream.
It accepts the following parameters:
Set dry gain. This sets input gain.
Set wet gain. This sets final output gain.
Set Impulse Response filter length. Default is 1, which means whole IR is processed.
This option is deprecated, and does nothing.
Set norm to be applied to IR coefficients before filtering. Allowed range is from -1 to 2. IR coefficients are normalized with calculated vector norm set by this option. For negative values, no norm is calculated, and IR coefficients are not modified at all. Default is 1.
For multichannel IR if this option is set to true, all IR channels will be
normalized with maximal measured gain of all IR channels coefficients as set by irnorm
option.
When disabled, all IR coefficients in each IR channel will be normalized independently.
Default is true.
Set gain to be applied to IR coefficients before filtering. Allowed range is 0 to 1. This gain is applied after any gain applied with irnorm option.
Set format of IR stream. Can be mono
or input
.
Default is input
.
Set max allowed Impulse Response filter duration in seconds. Default is 30 seconds. Allowed range is 0.1 to 60 seconds.
This option is deprecated, and does nothing.
This option is deprecated, and does nothing.
This option is deprecated, and does nothing.
This option is deprecated, and does nothing.
Set minimal partition size used for convolution. Default is 8192. Allowed range is from 1 to 65536. Lower values decreases latency at cost of higher CPU usage.
Set maximal partition size used for convolution. Default is 8192. Allowed range is from 8 to 65536. Lower values may increase CPU usage.
Set number of input impulse responses streams which will be switchable at runtime. Allowed range is from 1 to 32. Default is 1.
Set IR stream which will be used for convolution, starting from 0, should always be
lower than supplied value by nbirs
option. Default is 0.
This option can be changed at runtime via commands.
Set which precision to use when processing samples.
Auto pick internal sample format depending on other filters.
Always use single-floating point precision sample format.
Always use double-floating point precision sample format.
Default value is auto.
Set when to load IR stream. Can be init
or access
.
First one load and prepares all IRs on initialization, second one
once on first access of specific IR.
Default is init
.
ffmpeg -i input.wav -i middle_tunnel_1way_mono.wav -lavfi afir output.wav
"pan=4C|c0=FL|c1=FL|c2=FR|c3=FR[a];amovie=l_ir.wav[LIR];amovie=r_ir.wav[RIR];[LIR][RIR]amerge[ir];[a][ir]afir=irfmt=input:irnorm=1.2,pan=stereo|FL<c0+c2|FR<c1+c3"
irgain
explicitly set to estimated value and with irnorm
disabled:
"pan=4C|c0=FL|c1=FL|c2=FR|c3=FR[a];amovie=l_ir.wav[LIR];amovie=r_ir.wav[RIR];[LIR][RIR]amerge[ir];[a][ir]afir=irfmt=input:irgain=-5dB:irnom=-1,pan=stereo|FL<c0+c2|FR<c1+c3"
Set output format constraints for the input audio. The framework will negotiate the most appropriate format to minimize conversions.
It accepts the following parameters:
A ’|’-separated list of requested sample formats.
A ’|’-separated list of requested sample rates.
A ’|’-separated list of requested channel layouts.
See (ffmpeg-utils)the Channel Layout section in the ffmpeg-utils(1) manual for the required syntax.
If a parameter is omitted, all values are allowed.
Force the output to either unsigned 8-bit or signed 16-bit stereo
aformat=sample_fmts=u8|s16:channel_layouts=stereo
Apply frequency shift to input audio samples.
The filter accepts the following options:
Specify frequency shift. Allowed range is -INT_MAX to INT_MAX. Default value is 0.0.
Set output gain applied to final output. Allowed range is from 0.0 to 1.0. Default value is 1.0.
Set filter order used for filtering. Allowed range is from 1 to 16. Default value is 8.
This filter supports the all above options as commands.
Reduce broadband noise from input samples using Wavelets.
A description of the accepted options follows.
Set the noise sigma, allowed range is from 0 to 1. Default value is 0. This option controls strength of denoising applied to input samples. Most useful way to set this option is via decibels, eg. -45dB.
Set the number of wavelet levels of decomposition. Allowed range is from 1 to 12. Default value is 10. Setting this too low make denoising performance very poor.
Set wavelet type for decomposition of input frame. They are sorted by number of coefficients, from lowest to highest. More coefficients means worse filtering speed, but overall better quality. Available wavelets are:
Set percent of full denoising. Allowed range is from 0 to 100 percent. Default value is 85 percent or partial denoising.
If enabled, first input frame will be used as noise profile. If first frame samples contain non-noise performance will be very poor.
If enabled, input frames are analyzed for presence of noise. If noise is detected with high possibility then input frame profile will be used for processing following frames, until new noise frame is detected.
Set size of single frame in number of samples. Allowed range is from 512 to 65536. Default frame size is 8192 samples.
Set softness applied inside thresholding function. Allowed range is from 0 to 10. Default softness is 1.
This filter supports the all above options as commands.
A gate is mainly used to reduce lower parts of a signal. This kind of signal processing reduces disturbing noise between useful signals.
Gating is done by detecting the volume below a chosen level threshold and dividing it by the factor set with ratio. The bottom of the noise floor is set via range. Because an exact manipulation of the signal would cause distortion of the waveform the reduction can be levelled over time. This is done by setting attack and release.
attack determines how long the signal has to fall below the threshold before any reduction will occur and release sets the time the signal has to rise above the threshold to reduce the reduction again. Shorter signals than the chosen attack time will be left untouched.
Set input level before filtering. Default is 1. Allowed range is from 0.015625 to 64.
Set the mode of operation. Can be upward
or downward
.
Default is downward
. If set to upward
mode, higher parts of signal
will be amplified, expanding dynamic range in upward direction.
Otherwise, in case of downward
lower parts of signal will be reduced.
Set the level of gain reduction when the signal is below the threshold. Default is 0.06125. Allowed range is from 0 to 1. Setting this to 0 disables reduction and then filter behaves like expander.
If a signal rises above this level the gain reduction is released. Default is 0.125. Allowed range is from 0 to 1.
Set a ratio by which the signal is reduced. Default is 2. Allowed range is from 1 to 9000.
Amount of milliseconds the signal has to rise above the threshold before gain reduction stops. Default is 20 milliseconds. Allowed range is from 0.01 to 9000.
Amount of milliseconds the signal has to fall below the threshold before the reduction is increased again. Default is 250 milliseconds. Allowed range is from 0.01 to 9000.
Set amount of amplification of signal after processing. Default is 1. Allowed range is from 1 to 64.
Curve the sharp knee around the threshold to enter gain reduction more softly. Default is 2.828427125. Allowed range is from 1 to 8.
Choose if exact signal should be taken for detection or an RMS like one.
Default is rms
. Can be peak
or rms
.
Choose if the average level between all channels or the louder channel affects
the reduction.
Default is average
. Can be average
or maximum
.
This filter supports the all above options as commands.
Apply an arbitrary Infinite Impulse Response filter.
It accepts the following parameters:
Set B/numerator/zeros/reflection coefficients.
Set A/denominator/poles/ladder coefficients.
Set channels gains.
Set input gain.
Set output gain.
Set coefficients format.
lattice-ladder function
analog transfer function
digital transfer function
Z-plane zeros/poles, cartesian (default)
Z-plane zeros/poles, polar radians
Z-plane zeros/poles, polar degrees
S-plane zeros/poles
Set type of processing.
direct processing
serial processing
parallel processing
Set filtering precision.
double-precision floating-point (default)
single-precision floating-point
32-bit integers
16-bit integers
Normalize filter coefficients, by default is enabled. Enabling it will normalize magnitude response at DC to 0dB.
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Show IR frequency response, magnitude(magenta), phase(green) and group delay(yellow) in additional video stream. By default it is disabled.
Set for which IR channel to display frequency response. By default is first channel displayed. This option is used only when response is enabled.
Set video stream size. This option is used only when response is enabled.
Coefficients in tf
and sf
format are separated by spaces and are in ascending
order.
Coefficients in zp
format are separated by spaces and order of coefficients
doesn’t matter. Coefficients in zp
format are complex numbers with i
imaginary unit.
Different coefficients and gains can be provided for every channel, in such case use ’|’ to separate coefficients or gains. Last provided coefficients will be used for all remaining channels.
aiir=k=1:z=7.957584807809675810E-1 -2.575128568908332300 3.674839853930788710 -2.57512875289799137 7.957586296317130880E-1:p=1 -2.86950072432325953 3.63022088054647218 -2.28075678147272232 6.361362326477423500E-1:f=tf:r=d
zp
format:
aiir=k=0.79575848078096756:z=0.80918701+0.58773007i 0.80918701-0.58773007i 0.80884700+0.58784055i 0.80884700-0.58784055i:p=0.63892345+0.59951235i 0.63892345-0.59951235i 0.79582691+0.44198673i 0.79582691-0.44198673i:f=zp:r=s
aiir=z=1.3057 0 0 0:p=1.3057 2.3892 2.1860 1:f=sf:r=d
The limiter prevents an input signal from rising over a desired threshold. This limiter uses lookahead technology to prevent your signal from distorting. It means that there is a small delay after the signal is processed. Keep in mind that the delay it produces is the attack time you set.
The filter accepts the following options:
Set input gain. Default is 1.
Set output gain. Default is 1.
Don’t let signals above this level pass the limiter. Default is 1.
The limiter will reach its attenuation level in this amount of time in milliseconds. Default is 5 milliseconds.
Come back from limiting to attenuation 1.0 in this amount of milliseconds. Default is 50 milliseconds.
When gain reduction is always needed ASC takes care of releasing to an average reduction level rather than reaching a reduction of 0 in the release time.
Select how much the release time is affected by ASC, 0 means nearly no changes in release time while 1 produces higher release times.
Auto level output signal. Default is enabled. This normalizes audio back to 0dB if enabled.
Compensate the delay introduced by using the lookahead buffer set with attack parameter. Also flush the valid audio data in the lookahead buffer when the stream hits EOF.
Depending on picked setting it is recommended to upsample input 2x or 4x times with aresample before applying this filter.
Apply a two-pole all-pass filter with central frequency (in Hz) frequency, and filter-width width. An all-pass filter changes the audio’s frequency to phase relationship without changing its frequency to amplitude relationship.
The filter accepts the following options:
Set frequency in Hz.
Set method to specify band-width of filter.
Hz
Q-Factor
octave
slope
kHz
Specify the band-width of a filter in width_type units.
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Specify which channels to filter, by default all available are filtered.
Normalize biquad coefficients, by default is disabled. Enabling it will normalize magnitude response at DC to 0dB.
Set the filter order, can be 1 or 2. Default is 2.
Set transform type of IIR filter.
Set precision of filtering.
Pick automatic sample format depending on surround filters.
Always use signed 16-bit.
Always use signed 32-bit.
Always use float 32-bit.
Always use float 64-bit.
This filter supports the following commands:
Change allpass frequency. Syntax for the command is : "frequency"
Change allpass width_type. Syntax for the command is : "width_type"
Change allpass width. Syntax for the command is : "width"
Change allpass mix. Syntax for the command is : "mix"
Loop audio samples.
The filter accepts the following options:
Set the number of loops. Setting this value to -1 will result in infinite loops. Default is 0.
Set maximal number of samples. Default is 0.
Set first sample of loop. Default is 0.
Set the time of loop start in seconds.
Only used if option named start is set to -1
.
Merge two or more audio streams into a single multi-channel stream.
The filter accepts the following options:
Set the number of inputs. Default is 2.
If the channel layouts of the inputs are disjoint, and therefore compatible, the channel layout of the output will be set accordingly and the channels will be reordered as necessary. If the channel layouts of the inputs are not disjoint, the output will have all the channels of the first input then all the channels of the second input, in that order, and the channel layout of the output will be the default value corresponding to the total number of channels.
For example, if the first input is in 2.1 (FL+FR+LF) and the second input is FC+BL+BR, then the output will be in 5.1, with the channels in the following order: a1, a2, b1, a3, b2, b3 (a1 is the first channel of the first input, b1 is the first channel of the second input).
On the other hand, if both input are in stereo, the output channels will be in the default order: a1, a2, b1, b2, and the channel layout will be arbitrarily set to 4.0, which may or may not be the expected value.
All inputs must have the same sample rate, and format.
If inputs do not have the same duration, the output will stop with the shortest.
amovie=left.wav [l] ; amovie=right.mp3 [r] ; [l] [r] amerge
ffmpeg -i input.mkv -filter_complex "[0:1][0:2][0:3][0:4][0:5][0:6] amerge=inputs=6" -c:a pcm_s16le output.mkv
Mixes multiple audio inputs into a single output.
Note that this filter only supports float samples (the amerge and pan audio filters support many formats). If the amix input has integer samples then aresample will be automatically inserted to perform the conversion to float samples.
It accepts the following parameters:
The number of inputs. If unspecified, it defaults to 2.
How to determine the end-of-stream.
The duration of the longest input. (default)
The duration of the shortest input.
The duration of the first input.
The transition time, in seconds, for volume renormalization when an input stream ends. The default value is 2 seconds.
Specify weight of each input audio stream as a sequence of numbers separated by a space. If fewer weights are specified compared to number of inputs, the last weight is assigned to the remaining inputs. Default weight for each input is 1.
Always scale inputs instead of only doing summation of samples. Beware of heavy clipping if inputs are not normalized prior or after filtering by this filter if this option is disabled. By default is enabled.
ffmpeg -i INPUT1 -i INPUT2 -i INPUT3 -filter_complex amix=inputs=3:duration=first:dropout_transition=3 OUTPUT
ffmpeg -i VOCALS -i MUSIC -filter_complex amix=inputs=2:duration=longest:dropout_transition=0:weights="1 0.25":normalize=0 OUTPUT
This filter supports the following commands:
Syntax is same as option with same name.
Multiply first audio stream with second audio stream and store result in output audio stream. Multiplication is done by multiplying each sample from first stream with sample at same position from second stream.
With this element-wise multiplication one can create amplitude fades and amplitude modulations.
High-order parametric multiband equalizer for each channel.
It accepts the following parameters:
This option string is in format: "cchn f=cf w=w g=g t=f | ..." Each equalizer band is separated by ’|’.
Set channel number to which equalization will be applied. If input doesn’t have that channel the entry is ignored.
Set central frequency for band. If input doesn’t have that frequency the entry is ignored.
Set band width in Hertz.
Set band gain in dB.
Set filter type for band, optional, can be:
Butterworth, this is default.
Chebyshev type 1.
Chebyshev type 2.
With this option activated frequency response of anequalizer is displayed in video stream.
Set video stream size. Only useful if curves option is activated.
Set max gain that will be displayed. Only useful if curves option is activated. Setting this to a reasonable value makes it possible to display gain which is derived from neighbour bands which are too close to each other and thus produce higher gain when both are activated.
Set frequency scale used to draw frequency response in video output. Can be linear or logarithmic. Default is logarithmic.
Set color for each channel curve which is going to be displayed in video stream. This is list of color names separated by space or by ’|’. Unrecognised or missing colors will be replaced by white color.
anequalizer=c0 f=200 w=100 g=-10 t=1|c1 f=200 w=100 g=-10 t=1
This filter supports the following commands:
Alter existing filter parameters. Syntax for the commands is : "fN|f=freq|w=width|g=gain"
fN is existing filter number, starting from 0, if no such filter is available error is returned. freq set new frequency parameter. width set new width parameter in Hertz. gain set new gain parameter in dB.
Full filter invocation with asendcmd may look like this: asendcmd=c=’4.0 anequalizer change 0|f=200|w=50|g=1’,anequalizer=...
Reduce broadband noise in audio samples using Non-Local Means algorithm.
Each sample is adjusted by looking for other samples with similar contexts. This context similarity is defined by comparing their surrounding patches of size p. Patches are searched in an area of r around the sample.
The filter accepts the following options:
Set denoising strength. Allowed range is from 0.00001 to 10000. Default value is 0.00001.
Set patch radius duration. Allowed range is from 1 to 100 milliseconds. Default value is 2 milliseconds.
Set research radius duration. Allowed range is from 2 to 300 milliseconds. Default value is 6 milliseconds.
Set the output mode.
It accepts the following values:
Pass input unchanged.
Pass noise filtered out.
Pass only noise.
Default value is o.
Set smooth factor. Default value is 11. Allowed range is from 1 to 1000.
This filter supports the all above options as commands.
Apply Normalized Least-Mean-(Squares|Fourth) algorithm to the first audio stream using the second audio stream.
This adaptive filter is used to mimic a desired filter by finding the filter coefficients that relate to producing the least mean square of the error signal (difference between the desired, 2nd input audio stream and the actual signal, the 1st input audio stream).
A description of the accepted options follows.
Set filter order.
Set filter mu.
Set the filter eps.
Set the filter leakage.
It accepts the following values:
Pass the 1st input.
Pass the 2nd input.
Pass difference between desired, 2nd input and error signal estimate.
Pass difference between input, 1st input and error signal estimate.
Pass error signal estimated samples.
Default value is o.
Set which precision to use when processing samples.
Auto pick internal sample format depending on other filters.
Always use single-floating point precision sample format.
Always use double-floating point precision sample format.
asplit[a][b],[a]adelay=32S|32S[a],[b][a]anlms=order=128:leakage=0.0005:mu=.5:out_mode=o
This filter supports the same commands as options, excluding option order
.
Pass the audio source unchanged to the output.
Pad the end of an audio stream with silence.
This can be used together with ffmpeg
-shortest to
extend audio streams to the same length as the video stream.
A description of the accepted options follows.
Set silence packet size. Default value is 4096.
Set the number of samples of silence to add to the end. After the value is reached, the stream is terminated. This option is mutually exclusive with whole_len.
Set the minimum total number of samples in the output audio stream. If the value is longer than the input audio length, silence is added to the end, until the value is reached. This option is mutually exclusive with pad_len.
Specify the duration of samples of silence to add. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax. Used only if set to non-negative value.
Specify the minimum total duration in the output audio stream. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax. Used only if set to non-negative value. If the value is longer than the input audio length, silence is added to the end, until the value is reached. This option is mutually exclusive with pad_dur
If neither the pad_len nor the whole_len nor pad_dur nor whole_dur option is set, the filter will add silence to the end of the input stream indefinitely.
Note that for ffmpeg 4.4 and earlier a zero pad_dur or whole_dur also caused the filter to add silence indefinitely.
apad=pad_len=1024
apad=whole_len=10000
ffmpeg
to pad the audio input with silence, so that the
video stream will always result the shortest and will be converted
until the end in the output file when using the shortest
option:
ffmpeg -i VIDEO -i AUDIO -filter_complex "[1:0]apad" -shortest OUTPUT
Add a phasing effect to the input audio.
A phaser filter creates series of peaks and troughs in the frequency spectrum. The position of the peaks and troughs are modulated so that they vary over time, creating a sweeping effect.
A description of the accepted parameters follows.
Set input gain. Default is 0.4.
Set output gain. Default is 0.74
Set delay in milliseconds. Default is 3.0.
Set decay. Default is 0.4.
Set modulation speed in Hz. Default is 0.5.
Set modulation type. Default is triangular.
It accepts the following values:
Apply phase shift to input audio samples.
The filter accepts the following options:
Specify phase shift. Allowed range is from -1.0 to 1.0. Default value is 0.0.
Set output gain applied to final output. Allowed range is from 0.0 to 1.0. Default value is 1.0.
Set filter order used for filtering. Allowed range is from 1 to 16. Default value is 8.
This filter supports the all above options as commands.
Measure Audio Peak Signal-to-Noise Ratio.
This filter takes two audio streams for input, and outputs first audio stream. Results are in dB per channel at end of either input.
Apply Psychoacoustic clipper to input audio stream.
The filter accepts the following options:
Set input gain. By default it is 1. Range is [0.015625 - 64].
Set output gain. By default it is 1. Range is [0.015625 - 64].
Set the clipping start value. Default value is 0dBFS or 1.
Output only difference samples, useful to hear introduced distortions. By default is disabled.
Set strength of adaptive distortion applied. Default value is 0.5. Allowed range is from 0 to 1.
Set number of iterations of psychoacoustic clipper. Allowed range is from 1 to 20. Default value is 10.
Auto level output signal. Default is disabled. This normalizes audio back to 0dBFS if enabled.
This filter supports the all above options as commands.
Audio pulsator is something between an autopanner and a tremolo. But it can produce funny stereo effects as well. Pulsator changes the volume of the left and right channel based on a LFO (low frequency oscillator) with different waveforms and shifted phases. This filter have the ability to define an offset between left and right channel. An offset of 0 means that both LFO shapes match each other. The left and right channel are altered equally - a conventional tremolo. An offset of 50% means that the shape of the right channel is exactly shifted in phase (or moved backwards about half of the frequency) - pulsator acts as an autopanner. At 1 both curves match again. Every setting in between moves the phase shift gapless between all stages and produces some "bypassing" sounds with sine and triangle waveforms. The more you set the offset near 1 (starting from the 0.5) the faster the signal passes from the left to the right speaker.
The filter accepts the following options:
Set input gain. By default it is 1. Range is [0.015625 - 64].
Set output gain. By default it is 1. Range is [0.015625 - 64].
Set waveform shape the LFO will use. Can be one of: sine, triangle, square, sawup or sawdown. Default is sine.
Set modulation. Define how much of original signal is affected by the LFO.
Set left channel offset. Default is 0. Allowed range is [0 - 1].
Set right channel offset. Default is 0.5. Allowed range is [0 - 1].
Set pulse width. Default is 1. Allowed range is [0 - 2].
Set possible timing mode. Can be one of: bpm, ms or hz. Default is hz.
Set bpm. Default is 120. Allowed range is [30 - 300]. Only used if timing is set to bpm.
Set ms. Default is 500. Allowed range is [10 - 2000]. Only used if timing is set to ms.
Set frequency in Hz. Default is 2. Allowed range is [0.01 - 100]. Only used if timing is set to hz.
Resample the input audio to the specified parameters, using the libswresample library. If none are specified then the filter will automatically convert between its input and output.
This filter is also able to stretch/squeeze the audio data to make it match the timestamps or to inject silence / cut out audio to make it match the timestamps, do a combination of both or do neither.
The filter accepts the syntax [sample_rate:]resampler_options, where sample_rate expresses a sample rate and resampler_options is a list of key=value pairs, separated by ":". See the (ffmpeg-resampler)"Resampler Options" section in the ffmpeg-resampler(1) manual for the complete list of supported options.
aresample=44100
aresample=async=1000
Reverse an audio clip.
Warning: This filter requires memory to buffer the entire clip, so trimming is suggested.
atrim=end=5,areverse
Apply Recursive Least Squares algorithm to the first audio stream using the second audio stream.
This adaptive filter is used to mimic a desired filter by recursively finding the filter coefficients that relate to producing the minimal weighted linear least squares cost function of the error signal (difference between the desired, 2nd input audio stream and the actual signal, the 1st input audio stream).
A description of the accepted options follows.
Set the filter order.
Set the forgetting factor.
Set the coefficient to initialize internal covariance matrix.
Set the filter output samples. It accepts the following values:
Pass the 1st input.
Pass the 2nd input.
Pass difference between desired, 2nd input and error signal estimate.
Pass difference between input, 1st input and error signal estimate.
Pass error signal estimated samples.
Default value is o.
Set which precision to use when processing samples.
Auto pick internal sample format depending on other filters.
Always use single-floating point precision sample format.
Always use double-floating point precision sample format.
Reduce noise from speech using Recurrent Neural Networks.
This filter accepts the following options:
Set train model file to load. This option is always required.
Set how much to mix filtered samples into final output. Allowed range is from -1 to 1. Default value is 1. Negative values are special, they set how much to keep filtered noise in the final filter output. Set this option to -1 to hear actual noise removed from input signal.
This filter supports the all above options as commands.
Measure Audio Signal-to-Distortion Ratio.
This filter takes two audio streams for input, and outputs first audio stream. Results are in dB per channel at end of either input.
Set the number of samples per each output audio frame.
The last output packet may contain a different number of samples, as the filter will flush all the remaining samples when the input audio signals its end.
The filter accepts the following options:
Set the number of frames per each output audio frame. The number is intended as the number of samples per each channel. Default value is 1024.
If set to 1, the filter will pad the last audio frame with zeroes, so that the last frame will contain the same number of samples as the previous ones. Default value is 1.
For example, to set the number of per-frame samples to 1234 and disable padding for the last frame, use:
asetnsamples=n=1234:p=0
Set the sample rate without altering the PCM data. This will result in a change of speed and pitch.
The filter accepts the following options:
Set the output sample rate. Default is 44100 Hz.
Show a line containing various information for each input audio frame. The input audio is not modified.
The shown line contains a sequence of key/value pairs of the form key:value.
The following values are shown in the output:
The (sequential) number of the input frame, starting from 0.
The presentation timestamp of the input frame, in time base units; the time base depends on the filter input pad, and is usually 1/sample_rate.
The presentation timestamp of the input frame in seconds.
The sample format.
The channel layout.
The sample rate for the audio frame.
The number of samples (per channel) in the frame.
The Adler-32 checksum (printed in hexadecimal) of the audio data. For planar audio, the data is treated as if all the planes were concatenated.
A list of Adler-32 checksums for each data plane.
Measure Audio Scaled-Invariant Signal-to-Distortion Ratio.
This filter takes two audio streams for input, and outputs first audio stream. Results are in dB per channel at end of either input.
Apply audio soft clipping.
Soft clipping is a type of distortion effect where the amplitude of a signal is saturated along a smooth curve, rather than the abrupt shape of hard-clipping.
This filter accepts the following options:
Set type of soft-clipping.
It accepts the following values:
Set threshold from where to start clipping. Default value is 0dB or 1.
Set gain applied to output. Default value is 0dB or 1.
Set additional parameter which controls sigmoid function.
Set oversampling factor.
This filter supports the all above options as commands.
Display frequency domain statistical information about the audio channels. Statistics are calculated and stored as metadata for each audio channel and for each audio frame.
It accepts the following option:
Set the window length in samples. Default value is 2048. Allowed range is from 32 to 65536.
Set window function.
It accepts the following values:
Default is hann
.
Set window overlap. Allowed range is from 0
to 1
. Default value is 0.5
.
Select the parameters which are measured. The metadata keys can be used as flags, default is all which measures everything. none disables all measurement.
A list of each metadata key follows:
Automatic Speech Recognition
This filter uses PocketSphinx for speech recognition. To enable
compilation of this filter, you need to configure FFmpeg with
--enable-pocketsphinx
.
It accepts the following options:
Set sampling rate of input audio. Defaults is 16000
.
This need to match speech models, otherwise one will get poor results.
Set dictionary containing acoustic model files.
Set pronunciation dictionary.
Set language model file.
Set language model set.
Set which language model to use.
Set output for log messages.
The filter exports recognized speech as the frame metadata lavfi.asr.text
.
Display time domain statistical information about the audio channels. Statistics are calculated and displayed for each audio channel and, where applicable, an overall figure is also given.
It accepts the following option:
Short window length in seconds, used for peak and trough RMS measurement.
Default is 0.05
(50 milliseconds). Allowed range is [0 - 10]
.
Set metadata injection. All the metadata keys are prefixed with lavfi.astats.X
,
where X
is channel number starting from 1 or string Overall
. Default is
disabled.
Available keys for each channel are: Bit_depth Crest_factor DC_offset Dynamic_range Entropy Flat_factor Max_difference Max_level Mean_difference Min_difference Min_level Noise_floor Noise_floor_count Number_of_Infs Number_of_NaNs Number_of_denormals Peak_count Abs_Peak_count Peak_level RMS_difference RMS_peak RMS_trough Zero_crossings Zero_crossings_rate
and for Overall
:
Bit_depth
DC_offset
Entropy
Flat_factor
Max_difference
Max_level
Mean_difference
Min_difference
Min_level
Noise_floor
Noise_floor_count
Number_of_Infs
Number_of_NaNs
Number_of_denormals
Number_of_samples
Peak_count
Abs_Peak_count
Peak_level
RMS_difference
RMS_level
RMS_peak
RMS_trough
For example, a full key looks like lavfi.astats.1.DC_offset
or
lavfi.astats.Overall.Peak_count
.
Read below for the description of the keys.
Set the number of frames over which cumulative stats are calculated before being reset. Default is disabled.
Select the parameters which are measured per channel. The metadata keys can be used as flags, default is all which measures everything. none disables all per channel measurement.
Select the parameters which are measured overall. The metadata keys can be used as flags, default is all which measures everything. none disables all overall measurement.
A description of the measure keys follow:
no measures
all measures
overall bit depth of audio, i.e. number of bits used for each sample
standard ratio of peak to RMS level (note: not in dB)
mean amplitude displacement from zero
measured dynamic range of audio in dB
entropy measured across whole audio, entropy of value near 1.0 is typically measured for white noise
flatness (i.e. consecutive samples with the same value) of the signal at its peak levels (i.e. either Min_level or Max_level)
maximal difference between two consecutive samples
maximal sample level
mean difference between two consecutive samples, i.e. the average of each difference between two consecutive samples
minimal difference between two consecutive samples
minimal sample level
minimum local peak measured in dBFS over a short window
number of occasions (not the number of samples) that the signal attained Noise floor
number of samples with an infinite value
number of samples with a NaN (not a number) value
number of samples with a subnormal value
number of samples
number of occasions (not the number of samples) that the signal attained either Min_level or Max_level
number of occasions that the absolute samples taken from the signal attained max absolute value of Min_level and Max_level
standard peak level measured in dBFS
Root Mean Square difference between two consecutive samples
standard RMS level measured in dBFS
peak and trough values for RMS level measured over a short window, measured in dBFS.
number of points where the waveform crosses the zero level axis
rate of Zero crossings and number of audio samples
Boost subwoofer frequencies.
The filter accepts the following options:
Set dry gain, how much of original signal is kept. Allowed range is from 0 to 1. Default value is 1.0.
Set wet gain, how much of filtered signal is kept. Allowed range is from 0 to 1. Default value is 1.0.
Set max boost factor. Allowed range is from 1 to 12. Default value is 2.
Set delay line decay gain value. Allowed range is from 0 to 1. Default value is 0.0.
Set delay line feedback gain value. Allowed range is from 0 to 1. Default value is 0.9.
Set cutoff frequency in Hertz. Allowed range is 50 to 900. Default value is 100.
Set slope amount for cutoff frequency. Allowed range is 0.0001 to 1. Default value is 0.5.
Set delay. Allowed range is from 1 to 100. Default value is 20.
Set the channels to process. Default value is all available.
This filter supports the all above options as commands.
Cut subwoofer frequencies.
This filter allows to set custom, steeper roll off than highpass filter, and thus is able to more attenuate frequency content in stop-band.
The filter accepts the following options:
Set cutoff frequency in Hertz. Allowed range is 2 to 200. Default value is 20.
Set filter order. Available values are from 3 to 20. Default value is 10.
Set input gain level. Allowed range is from 0 to 1. Default value is 1.
This filter supports the all above options as commands.
Cut super frequencies.
The filter accepts the following options:
Set cutoff frequency in Hertz. Allowed range is 20000 to 192000. Default value is 20000.
Set filter order. Available values are from 3 to 20. Default value is 10.
Set input gain level. Allowed range is from 0 to 1. Default value is 1.
This filter supports the all above options as commands.
Apply high order Butterworth band-pass filter.
The filter accepts the following options:
Set center frequency in Hertz. Allowed range is 2 to 999999. Default value is 1000.
Set filter order. Available values are from 4 to 20. Default value is 4.
Set Q-factor. Allowed range is from 0.01 to 100. Default value is 1.
Set input gain level. Allowed range is from 0 to 2. Default value is 1.
This filter supports the all above options as commands.
Apply high order Butterworth band-stop filter.
The filter accepts the following options:
Set center frequency in Hertz. Allowed range is 2 to 999999. Default value is 1000.
Set filter order. Available values are from 4 to 20. Default value is 4.
Set Q-factor. Allowed range is from 0.01 to 100. Default value is 1.
Set input gain level. Allowed range is from 0 to 2. Default value is 1.
This filter supports the all above options as commands.
Adjust audio tempo.
The filter accepts exactly one parameter, the audio tempo. If not specified then the filter will assume nominal 1.0 tempo. Tempo must be in the [0.5, 100.0] range.
Note that tempo greater than 2 will skip some samples rather than blend them in. If for any reason this is a concern it is always possible to daisy-chain several instances of atempo to achieve the desired product tempo.
atempo=0.8
atempo=3
atempo=sqrt(3),atempo=sqrt(3)
This filter supports the following commands:
Change filter tempo scale factor. Syntax for the command is : "tempo"
Apply spectral tilt filter to audio stream.
This filter apply any spectral roll-off slope over any specified frequency band.
The filter accepts the following options:
Set central frequency of tilt in Hz. Default is 10000 Hz.
Set slope direction of tilt. Default is 0. Allowed range is from -1 to 1.
Set width of tilt. Default is 1000. Allowed range is from 100 to 10000.
Set order of tilt filter.
Set input volume level. Allowed range is from 0 to 4. Default is 1.
This filter supports the all above options as commands.
Trim the input so that the output contains one continuous subpart of the input.
It accepts the following parameters:
Timestamp (in seconds) of the start of the section to keep. I.e. the audio sample with the timestamp start will be the first sample in the output.
Specify time of the first audio sample that will be dropped, i.e. the audio sample immediately preceding the one with the timestamp end will be the last sample in the output.
Same as start, except this option sets the start timestamp in samples instead of seconds.
Same as end, except this option sets the end timestamp in samples instead of seconds.
The maximum duration of the output in seconds.
The number of the first sample that should be output.
The number of the first sample that should be dropped.
start, end, and duration are expressed as time duration specifications; see (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual.
Note that the first two sets of the start/end options and the duration option look at the frame timestamp, while the _sample options simply count the samples that pass through the filter. So start/end_pts and start/end_sample will give different results when the timestamps are wrong, inexact or do not start at zero. Also note that this filter does not modify the timestamps. If you wish to have the output timestamps start at zero, insert the asetpts filter after the atrim filter.
If multiple start or end options are set, this filter tries to be greedy and keep all samples that match at least one of the specified constraints. To keep only the part that matches all the constraints at once, chain multiple atrim filters.
The defaults are such that all the input is kept. So it is possible to set e.g. just the end values to keep everything before the specified time.
Examples:
ffmpeg -i INPUT -af atrim=60:120
ffmpeg -i INPUT -af atrim=end_sample=1000
Calculate normalized windowed cross-correlation between two input audio streams.
Resulted samples are always between -1 and 1 inclusive. If result is 1 it means two input samples are highly correlated in that selected segment. Result 0 means they are not correlated at all. If result is -1 it means two input samples are out of phase, which means they cancel each other.
The filter accepts the following options:
Set size of segment over which cross-correlation is calculated. Default is 256. Allowed range is from 2 to 131072.
Set algorithm for cross-correlation. Can be slow
or fast
or best
.
Default is best
. Fast algorithm assumes mean values over any given segment
are always zero and thus need much less calculations to make.
This is generally not true, but is valid for typical audio streams.
ffmpeg -i stereo.wav -af channelsplit,axcorrelate=size=1024:algo=fast correlation.wav
Apply a two-pole Butterworth band-pass filter with central frequency frequency, and (3dB-point) band-width width. The csg option selects a constant skirt gain (peak gain = Q) instead of the default: constant 0dB peak gain. The filter roll off at 6dB per octave (20dB per decade).
The filter accepts the following options:
Set the filter’s central frequency. Default is 3000
.
Constant skirt gain if set to 1. Defaults to 0.
Set method to specify band-width of filter.
Hz
Q-Factor
octave
slope
kHz
Specify the band-width of a filter in width_type units.
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Specify which channels to filter, by default all available are filtered.
Normalize biquad coefficients, by default is disabled. Enabling it will normalize magnitude response at DC to 0dB.
Set transform type of IIR filter.
Set precision of filtering.
Pick automatic sample format depending on surround filters.
Always use signed 16-bit.
Always use signed 32-bit.
Always use float 32-bit.
Always use float 64-bit.
Set block size used for reverse IIR processing. If this value is set to high enough value (higher than impulse response length truncated when reaches near zero values) filtering will become linear phase otherwise if not big enough it will just produce nasty artifacts.
Note that filter delay will be exactly this many samples when set to non-zero value.
This filter supports the following commands:
Change bandpass frequency. Syntax for the command is : "frequency"
Change bandpass width_type. Syntax for the command is : "width_type"
Change bandpass width. Syntax for the command is : "width"
Change bandpass mix. Syntax for the command is : "mix"
Apply a two-pole Butterworth band-reject filter with central frequency frequency, and (3dB-point) band-width width. The filter roll off at 6dB per octave (20dB per decade).
The filter accepts the following options:
Set the filter’s central frequency. Default is 3000
.
Set method to specify band-width of filter.
Hz
Q-Factor
octave
slope
kHz
Specify the band-width of a filter in width_type units.
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Specify which channels to filter, by default all available are filtered.
Normalize biquad coefficients, by default is disabled. Enabling it will normalize magnitude response at DC to 0dB.
Set transform type of IIR filter.
Set precision of filtering.
Pick automatic sample format depending on surround filters.
Always use signed 16-bit.
Always use signed 32-bit.
Always use float 32-bit.
Always use float 64-bit.
Set block size used for reverse IIR processing. If this value is set to high enough value (higher than impulse response length truncated when reaches near zero values) filtering will become linear phase otherwise if not big enough it will just produce nasty artifacts.
Note that filter delay will be exactly this many samples when set to non-zero value.
This filter supports the following commands:
Change bandreject frequency. Syntax for the command is : "frequency"
Change bandreject width_type. Syntax for the command is : "width_type"
Change bandreject width. Syntax for the command is : "width"
Change bandreject mix. Syntax for the command is : "mix"
Boost or cut the bass (lower) frequencies of the audio using a two-pole shelving filter with a response similar to that of a standard hi-fi’s tone-controls. This is also known as shelving equalisation (EQ).
The filter accepts the following options:
Give the gain at 0 Hz. Its useful range is about -20 (for a large cut) to +20 (for a large boost). Beware of clipping when using a positive gain.
Set the filter’s central frequency and so can be used
to extend or reduce the frequency range to be boosted or cut.
The default value is 100
Hz.
Set method to specify band-width of filter.
Hz
Q-Factor
octave
slope
kHz
Determine how steep is the filter’s shelf transition.
Set number of poles. Default is 2.
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Specify which channels to filter, by default all available are filtered.
Normalize biquad coefficients, by default is disabled. Enabling it will normalize magnitude response at DC to 0dB.
Set transform type of IIR filter.
Set precision of filtering.
Pick automatic sample format depending on surround filters.
Always use signed 16-bit.
Always use signed 32-bit.
Always use float 32-bit.
Always use float 64-bit.
Set block size used for reverse IIR processing. If this value is set to high enough value (higher than impulse response length truncated when reaches near zero values) filtering will become linear phase otherwise if not big enough it will just produce nasty artifacts.
Note that filter delay will be exactly this many samples when set to non-zero value.
This filter supports the following commands:
Change bass frequency. Syntax for the command is : "frequency"
Change bass width_type. Syntax for the command is : "width_type"
Change bass width. Syntax for the command is : "width"
Change bass gain. Syntax for the command is : "gain"
Change bass mix. Syntax for the command is : "mix"
Apply a biquad IIR filter with the given coefficients. Where b0, b1, b2 and a0, a1, a2 are the numerator and denominator coefficients respectively. and channels, c specify which channels to filter, by default all available are filtered.
This filter supports the following commands:
Change biquad parameter. Syntax for the command is : "value"
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Specify which channels to filter, by default all available are filtered.
Normalize biquad coefficients, by default is disabled. Enabling it will normalize magnitude response at DC to 0dB.
Set transform type of IIR filter.
Set precision of filtering.
Pick automatic sample format depending on surround filters.
Always use signed 16-bit.
Always use signed 32-bit.
Always use float 32-bit.
Always use float 64-bit.
Set block size used for reverse IIR processing. If this value is set to high enough value (higher than impulse response length truncated when reaches near zero values) filtering will become linear phase otherwise if not big enough it will just produce nasty artifacts.
Note that filter delay will be exactly this many samples when set to non-zero value.
Bauer stereo to binaural transformation, which improves headphone listening of stereo audio records.
To enable compilation of this filter you need to configure FFmpeg with
--enable-libbs2b
.
It accepts the following parameters:
Pre-defined crossfeed level.
Default level (fcut=700, feed=50).
Chu Moy circuit (fcut=700, feed=60).
Jan Meier circuit (fcut=650, feed=95).
Cut frequency (in Hz).
Feed level (in Hz).
Remap input channels to new locations.
It accepts the following parameters:
Map channels from input to output. The argument is a ’|’-separated list of
mappings, each in the in_channel-out_channel
or
in_channel form. in_channel can be either the name of the input
channel (e.g. FL for front left) or its index in the input channel layout.
out_channel is the name of the output channel or its index in the output
channel layout. If out_channel is not given then it is implicitly an
index, starting with zero and increasing by one for each mapping.
The channel layout of the output stream.
If no mapping is present, the filter will implicitly map input channels to output channels, preserving indices.
ffmpeg -i in.mov -filter 'channelmap=map=DL-FL|DR-FR' out.wav
will create an output WAV file tagged as stereo from the downmix channels of the input.
ffmpeg -i in.wav -filter 'channelmap=1|2|0|5|3|4:5.1' out.wav
Split each channel from an input audio stream into a separate output stream.
It accepts the following parameters:
The channel layout of the input stream. The default is "stereo".
A channel layout describing the channels to be extracted as separate output streams or "all" to extract each input channel as a separate stream. The default is "all".
Choosing channels not present in channel layout in the input will result in an error.
ffmpeg -i in.mp3 -filter_complex channelsplit out.mkv
will create an output Matroska file with two audio streams, one containing only the left channel and the other the right channel.
ffmpeg -i in.wav -filter_complex 'channelsplit=channel_layout=5.1[FL][FR][FC][LFE][SL][SR]' -map '[FL]' front_left.wav -map '[FR]' front_right.wav -map '[FC]' front_center.wav -map '[LFE]' lfe.wav -map '[SL]' side_left.wav -map '[SR]' side_right.wav
ffmpeg -i in.wav -filter_complex 'channelsplit=channel_layout=5.1:channels=LFE[LFE]' -map '[LFE]' lfe.wav
Add a chorus effect to the audio.
Can make a single vocal sound like a chorus, but can also be applied to instrumentation.
Chorus resembles an echo effect with a short delay, but whereas with echo the delay is constant, with chorus, it is varied using using sinusoidal or triangular modulation. The modulation depth defines the range the modulated delay is played before or after the delay. Hence the delayed sound will sound slower or faster, that is the delayed sound tuned around the original one, like in a chorus where some vocals are slightly off key.
It accepts the following parameters:
Set input gain. Default is 0.4.
Set output gain. Default is 0.4.
Set delays. A typical delay is around 40ms to 60ms.
Set decays.
Set speeds.
Set depths.
chorus=0.7:0.9:55:0.4:0.25:2
chorus=0.6:0.9:50|60:0.4|0.32:0.25|0.4:2|1.3
chorus=0.5:0.9:50|60|40:0.4|0.32|0.3:0.25|0.4|0.3:2|2.3|1.3
Compress or expand the audio’s dynamic range.
It accepts the following parameters:
A list of times in seconds for each channel over which the instantaneous level of the input signal is averaged to determine its volume. attacks refers to increase of volume and decays refers to decrease of volume. For most situations, the attack time (response to the audio getting louder) should be shorter than the decay time, because the human ear is more sensitive to sudden loud audio than sudden soft audio. A typical value for attack is 0.3 seconds and a typical value for decay is 0.8 seconds. If specified number of attacks & decays is lower than number of channels, the last set attack/decay will be used for all remaining channels.
A list of points for the transfer function, specified in dB relative to the
maximum possible signal amplitude. Each key points list must be defined using
the following syntax: x0/y0|x1/y1|x2/y2|....
or
x0/y0 x1/y1 x2/y2 ....
The input values must be in strictly increasing order but the transfer function
does not have to be monotonically rising. The point 0/0
is assumed but
may be overridden (by 0/out-dBn
). Typical values for the transfer
function are -70/-70|-60/-20|1/0
.
Set the curve radius in dB for all joints. It defaults to 0.01.
Set the additional gain in dB to be applied at all points on the transfer function. This allows for easy adjustment of the overall gain. It defaults to 0.
Set an initial volume, in dB, to be assumed for each channel when filtering starts. This permits the user to supply a nominal level initially, so that, for example, a very large gain is not applied to initial signal levels before the companding has begun to operate. A typical value for audio which is initially quiet is -90 dB. It defaults to 0.
Set a delay, in seconds. The input audio is analyzed immediately, but audio is delayed before being fed to the volume adjuster. Specifying a delay approximately equal to the attack/decay times allows the filter to effectively operate in predictive rather than reactive mode. It defaults to 0.
compand=.3|.3:1|1:-90/-60|-60/-40|-40/-30|-20/-20:6:0:-90:0.2
Another example for audio with whisper and explosion parts:
compand=0|0:1|1:-90/-900|-70/-70|-30/-9|0/-3:6:0:0:0
compand=.1|.1:.2|.2:-900/-900|-50.1/-900|-50/-50:.01:0:-90:.1
compand=.1|.1:.1|.1:-45.1/-45.1|-45/-900|0/-900:.01:45:-90:.1
compand=points=-80/-80|-6/-6|0/-3.8|20/3.5
compand=points=-80/-80|-9/-9|0/-5.3|20/2.9
compand=points=-80/-80|-12/-12|0/-6.8|20/1.9
compand=points=-80/-80|-18/-18|0/-9.8|20/0.7
compand=points=-80/-80|-15/-15|0/-10.8|20/-5.2
compand=points=-80/-105|-62/-80|-15.4/-15.4|0/-12|20/-7.6
compand=attacks=0:points=-80/-169|-54/-80|-49.5/-64.6|-41.1/-41.1|-25.8/-15|-10.8/-4.5|0/0|20/8.3
compand=attacks=0:points=-80/-80|-6/-6|20/-6
compand=attacks=0:points=-80/-80|-12/-12|20/-12
compand=attacks=0:points=-80/-115|-35.1/-80|-35/-35|20/20
compand=attacks=0:points=-80/-80|-12.4/-12.4|-6/-8|0/-6.8|20/-2.8
Compensation Delay Line is a metric based delay to compensate differing positions of microphones or speakers.
For example, you have recorded guitar with two microphones placed in different locations. Because the front of sound wave has fixed speed in normal conditions, the phasing of microphones can vary and depends on their location and interposition. The best sound mix can be achieved when these microphones are in phase (synchronized). Note that a distance of ~30 cm between microphones makes one microphone capture the signal in antiphase to the other microphone. That makes the final mix sound moody. This filter helps to solve phasing problems by adding different delays to each microphone track and make them synchronized.
The best result can be reached when you take one track as base and synchronize other tracks one by one with it. Remember that synchronization/delay tolerance depends on sample rate, too. Higher sample rates will give more tolerance.
The filter accepts the following parameters:
Set millimeters distance. This is compensation distance for fine tuning. Default is 0.
Set cm distance. This is compensation distance for tightening distance setup. Default is 0.
Set meters distance. This is compensation distance for hard distance setup. Default is 0.
Set dry amount. Amount of unprocessed (dry) signal. Default is 0.
Set wet amount. Amount of processed (wet) signal. Default is 1.
Set temperature in degrees Celsius. This is the temperature of the environment. Default is 20.
This filter supports the all above options as commands.
Apply headphone crossfeed filter.
Crossfeed is the process of blending the left and right channels of stereo audio recording. It is mainly used to reduce extreme stereo separation of low frequencies.
The intent is to produce more speaker like sound to the listener.
The filter accepts the following options:
Set strength of crossfeed. Default is 0.2. Allowed range is from 0 to 1. This sets gain of low shelf filter for side part of stereo image. Default is -6dB. Max allowed is -30db when strength is set to 1.
Set soundstage wideness. Default is 0.5. Allowed range is from 0 to 1. This sets cut off frequency of low shelf filter. Default is cut off near 1550 Hz. With range set to 1 cut off frequency is set to 2100 Hz.
Set curve slope of low shelf filter. Default is 0.5. Allowed range is from 0.01 to 1.
Set input gain. Default is 0.9.
Set output gain. Default is 1.
Set block size used for reverse IIR processing. If this value is set to high enough value (higher than impulse response length truncated when reaches near zero values) filtering will become linear phase otherwise if not big enough it will just produce nasty artifacts.
Note that filter delay will be exactly this many samples when set to non-zero value.
This filter supports the all above options as commands.
Simple algorithm for audio noise sharpening.
This filter linearly increases differences between each audio sample.
The filter accepts the following options:
Sets the intensity of effect (default: 2.0). Must be in range between -10.0 to 0 (unchanged sound) to 10.0 (maximum effect). To inverse filtering use negative value.
Enable clipping. By default is enabled.
This filter supports the all above options as commands.
Apply a DC shift to the audio.
This can be useful to remove a DC offset (caused perhaps by a hardware problem in the recording chain) from the audio. The effect of a DC offset is reduced headroom and hence volume. The astats filter can be used to determine if a signal has a DC offset.
Set the DC shift, allowed range is [-1, 1]. It indicates the amount to shift the audio.
Optional. It should have a value much less than 1 (e.g. 0.05 or 0.02) and is used to prevent clipping.
Apply de-essing to the audio samples.
Set intensity for triggering de-essing. Allowed range is from 0 to 1. Default is 0.
Set amount of ducking on treble part of sound. Allowed range is from 0 to 1. Default is 0.5.
How much of original frequency content to keep when de-essing. Allowed range is from 0 to 1. Default is 0.5.
Set the output mode.
It accepts the following values:
Pass input unchanged.
Pass ess filtered out.
Pass only ess.
Default value is o.
Enhance dialogue in stereo audio.
This filter accepts stereo input and produce surround (3.0) channels output. The newly produced front center channel have enhanced speech dialogue originally available in both stereo channels. This filter outputs front left and front right channels same as available in stereo input.
The filter accepts the following options:
Set the original center factor to keep in front center channel output. Allowed range is from 0 to 1. Default value is 1.
Set the dialogue enhance factor to put in front center channel output. Allowed range is from 0 to 3. Default value is 1.
Set the voice detection factor. Allowed range is from 2 to 32. Default value is 2.
This filter supports the all above options as commands.
Measure audio dynamic range.
DR values of 14 and higher is found in very dynamic material. DR of 8 to 13 is found in transition material. And anything less that 8 have very poor dynamics and is very compressed.
The filter accepts the following options:
Set window length in seconds used to split audio into segments of equal length. Default is 3 seconds.
Dynamic Audio Normalizer.
This filter applies a certain amount of gain to the input audio in order to bring its peak magnitude to a target level (e.g. 0 dBFS). However, in contrast to more "simple" normalization algorithms, the Dynamic Audio Normalizer *dynamically* re-adjusts the gain factor to the input audio. This allows for applying extra gain to the "quiet" sections of the audio while avoiding distortions or clipping the "loud" sections. In other words: The Dynamic Audio Normalizer will "even out" the volume of quiet and loud sections, in the sense that the volume of each section is brought to the same target level. Note, however, that the Dynamic Audio Normalizer achieves this goal *without* applying "dynamic range compressing". It will retain 100% of the dynamic range *within* each section of the audio file.
Set the frame length in milliseconds. In range from 10 to 8000 milliseconds. Default is 500 milliseconds. The Dynamic Audio Normalizer processes the input audio in small chunks, referred to as frames. This is required, because a peak magnitude has no meaning for just a single sample value. Instead, we need to determine the peak magnitude for a contiguous sequence of sample values. While a "standard" normalizer would simply use the peak magnitude of the complete file, the Dynamic Audio Normalizer determines the peak magnitude individually for each frame. The length of a frame is specified in milliseconds. By default, the Dynamic Audio Normalizer uses a frame length of 500 milliseconds, which has been found to give good results with most files. Note that the exact frame length, in number of samples, will be determined automatically, based on the sampling rate of the individual input audio file.
Set the Gaussian filter window size. In range from 3 to 301, must be odd
number. Default is 31.
Probably the most important parameter of the Dynamic Audio Normalizer is the
window size
of the Gaussian smoothing filter. The filter’s window size
is specified in frames, centered around the current frame. For the sake of
simplicity, this must be an odd number. Consequently, the default value of 31
takes into account the current frame, as well as the 15 preceding frames and
the 15 subsequent frames. Using a larger window results in a stronger
smoothing effect and thus in less gain variation, i.e. slower gain
adaptation. Conversely, using a smaller window results in a weaker smoothing
effect and thus in more gain variation, i.e. faster gain adaptation.
In other words, the more you increase this value, the more the Dynamic Audio
Normalizer will behave like a "traditional" normalization filter. On the
contrary, the more you decrease this value, the more the Dynamic Audio
Normalizer will behave like a dynamic range compressor.
Set the target peak value. This specifies the highest permissible magnitude level for the normalized audio input. This filter will try to approach the target peak magnitude as closely as possible, but at the same time it also makes sure that the normalized signal will never exceed the peak magnitude. A frame’s maximum local gain factor is imposed directly by the target peak magnitude. The default value is 0.95 and thus leaves a headroom of 5%*. It is not recommended to go above this value.
Set the maximum gain factor. In range from 1.0 to 100.0. Default is 10.0. The Dynamic Audio Normalizer determines the maximum possible (local) gain factor for each input frame, i.e. the maximum gain factor that does not result in clipping or distortion. The maximum gain factor is determined by the frame’s highest magnitude sample. However, the Dynamic Audio Normalizer additionally bounds the frame’s maximum gain factor by a predetermined (global) maximum gain factor. This is done in order to avoid excessive gain factors in "silent" or almost silent frames. By default, the maximum gain factor is 10.0, For most inputs the default value should be sufficient and it usually is not recommended to increase this value. Though, for input with an extremely low overall volume level, it may be necessary to allow even higher gain factors. Note, however, that the Dynamic Audio Normalizer does not simply apply a "hard" threshold (i.e. cut off values above the threshold). Instead, a "sigmoid" threshold function will be applied. This way, the gain factors will smoothly approach the threshold value, but never exceed that value.
Set the target RMS. In range from 0.0 to 1.0. Default is 0.0 - disabled. By default, the Dynamic Audio Normalizer performs "peak" normalization. This means that the maximum local gain factor for each frame is defined (only) by the frame’s highest magnitude sample. This way, the samples can be amplified as much as possible without exceeding the maximum signal level, i.e. without clipping. Optionally, however, the Dynamic Audio Normalizer can also take into account the frame’s root mean square, abbreviated RMS. In electrical engineering, the RMS is commonly used to determine the power of a time-varying signal. It is therefore considered that the RMS is a better approximation of the "perceived loudness" than just looking at the signal’s peak magnitude. Consequently, by adjusting all frames to a constant RMS value, a uniform "perceived loudness" can be established. If a target RMS value has been specified, a frame’s local gain factor is defined as the factor that would result in exactly that RMS value. Note, however, that the maximum local gain factor is still restricted by the frame’s highest magnitude sample, in order to prevent clipping.
Enable channels coupling. By default is enabled. By default, the Dynamic Audio Normalizer will amplify all channels by the same amount. This means the same gain factor will be applied to all channels, i.e. the maximum possible gain factor is determined by the "loudest" channel. However, in some recordings, it may happen that the volume of the different channels is uneven, e.g. one channel may be "quieter" than the other one(s). In this case, this option can be used to disable the channel coupling. This way, the gain factor will be determined independently for each channel, depending only on the individual channel’s highest magnitude sample. This allows for harmonizing the volume of the different channels.
Enable DC bias correction. By default is disabled. An audio signal (in the time domain) is a sequence of sample values. In the Dynamic Audio Normalizer these sample values are represented in the -1.0 to 1.0 range, regardless of the original input format. Normally, the audio signal, or "waveform", should be centered around the zero point. That means if we calculate the mean value of all samples in a file, or in a single frame, then the result should be 0.0 or at least very close to that value. If, however, there is a significant deviation of the mean value from 0.0, in either positive or negative direction, this is referred to as a DC bias or DC offset. Since a DC bias is clearly undesirable, the Dynamic Audio Normalizer provides optional DC bias correction. With DC bias correction enabled, the Dynamic Audio Normalizer will determine the mean value, or "DC correction" offset, of each input frame and subtract that value from all of the frame’s sample values which ensures those samples are centered around 0.0 again. Also, in order to avoid "gaps" at the frame boundaries, the DC correction offset values will be interpolated smoothly between neighbouring frames.
Enable alternative boundary mode. By default is disabled. The Dynamic Audio Normalizer takes into account a certain neighbourhood around each frame. This includes the preceding frames as well as the subsequent frames. However, for the "boundary" frames, located at the very beginning and at the very end of the audio file, not all neighbouring frames are available. In particular, for the first few frames in the audio file, the preceding frames are not known. And, similarly, for the last few frames in the audio file, the subsequent frames are not known. Thus, the question arises which gain factors should be assumed for the missing frames in the "boundary" region. The Dynamic Audio Normalizer implements two modes to deal with this situation. The default boundary mode assumes a gain factor of exactly 1.0 for the missing frames, resulting in a smooth "fade in" and "fade out" at the beginning and at the end of the input, respectively.
Set the compress factor. In range from 0.0 to 30.0. Default is 0.0. By default, the Dynamic Audio Normalizer does not apply "traditional" compression. This means that signal peaks will not be pruned and thus the full dynamic range will be retained within each local neighbourhood. However, in some cases it may be desirable to combine the Dynamic Audio Normalizer’s normalization algorithm with a more "traditional" compression. For this purpose, the Dynamic Audio Normalizer provides an optional compression (thresholding) function. If (and only if) the compression feature is enabled, all input frames will be processed by a soft knee thresholding function prior to the actual normalization process. Put simply, the thresholding function is going to prune all samples whose magnitude exceeds a certain threshold value. However, the Dynamic Audio Normalizer does not simply apply a fixed threshold value. Instead, the threshold value will be adjusted for each individual frame. In general, smaller parameters result in stronger compression, and vice versa. Values below 3.0 are not recommended, because audible distortion may appear.
Set the target threshold value. This specifies the lowest permissible magnitude level for the audio input which will be normalized. If input frame volume is above this value frame will be normalized. Otherwise frame may not be normalized at all. The default value is set to 0, which means all input frames will be normalized. This option is mostly useful if digital noise is not wanted to be amplified.
Specify which channels to filter, by default all available channels are filtered.
Specify overlap for frames. If set to 0 (default) no frame overlapping is done. Using >0 and <1 values will make less conservative gain adjustments, like when framelen option is set to smaller value, if framelen option value is compensated for non-zero overlap then gain adjustments will be smoother across time compared to zero overlap case.
Specify the peak mapping curve expression which is going to be used when calculating gain applied to frames. The max output frame gain will still be limited by other options mentioned previously for this filter.
The expression can contain the following constants:
current channel number
current sample number
number of channels
timestamp expressed in seconds
sample rate
current frame peak value
This filter supports the all above options as commands.
Make audio easier to listen to on headphones.
This filter adds ‘cues’ to 44.1kHz stereo (i.e. audio CD format) audio so that when listened to on headphones the stereo image is moved from inside your head (standard for headphones) to outside and in front of the listener (standard for speakers).
Ported from SoX.
Apply a two-pole peaking equalisation (EQ) filter. With this filter, the signal-level at and around a selected frequency can be increased or decreased, whilst (unlike bandpass and bandreject filters) that at all other frequencies is unchanged.
In order to produce complex equalisation curves, this filter can be given several times, each with a different central frequency.
The filter accepts the following options:
Set the filter’s central frequency in Hz.
Set method to specify band-width of filter.
Hz
Q-Factor
octave
slope
kHz
Specify the band-width of a filter in width_type units.
Set the required gain or attenuation in dB. Beware of clipping when using a positive gain.
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Specify which channels to filter, by default all available are filtered.
Normalize biquad coefficients, by default is disabled. Enabling it will normalize magnitude response at DC to 0dB.
Set transform type of IIR filter.
Set precision of filtering.
Pick automatic sample format depending on surround filters.
Always use signed 16-bit.
Always use signed 32-bit.
Always use float 32-bit.
Always use float 64-bit.
Set block size used for reverse IIR processing. If this value is set to high enough value (higher than impulse response length truncated when reaches near zero values) filtering will become linear phase otherwise if not big enough it will just produce nasty artifacts.
Note that filter delay will be exactly this many samples when set to non-zero value.
equalizer=f=1000:t=h:width=200:g=-10
equalizer=f=1000:t=q:w=1:g=2,equalizer=f=100:t=q:w=2:g=-5
This filter supports the following commands:
Change equalizer frequency. Syntax for the command is : "frequency"
Change equalizer width_type. Syntax for the command is : "width_type"
Change equalizer width. Syntax for the command is : "width"
Change equalizer gain. Syntax for the command is : "gain"
Change equalizer mix. Syntax for the command is : "mix"
Linearly increases the difference between left and right channels which adds some sort of "live" effect to playback.
The filter accepts the following options:
Sets the difference coefficient (default: 2.5). 0.0 means mono sound (average of both channels), with 1.0 sound will be unchanged, with -1.0 left and right channels will be swapped.
Enable clipping. By default is enabled.
This filter supports the all above options as commands.
Apply FIR Equalization using arbitrary frequency response.
The filter accepts the following option:
Set gain curve equation (in dB). The expression can contain variables:
the evaluated frequency
sample rate
channel number, set to 0 when multichannels evaluation is disabled
channel id, see libavutil/channel_layout.h, set to the first channel id when multichannels evaluation is disabled
number of channels
channel_layout, see libavutil/channel_layout.h
and functions:
interpolate gain on frequency f based on gain_entry
same as gain_interpolate, but smoother
This option is also available as command. Default is gain_interpolate(f)
.
Set gain entry for gain_interpolate function. The expression can contain functions:
store gain entry at frequency f with value g
This option is also available as command.
Set filter delay in seconds. Higher value means more accurate.
Default is 0.01
.
Set filter accuracy in Hz. Lower value means more accurate.
Default is 5
.
Set window function. Acceptable values are:
rectangular window, useful when gain curve is already smooth
hann window (default)
hamming window
blackman window
3-terms continuous 1st derivative nuttall window
minimum 3-terms discontinuous nuttall window
4-terms continuous 1st derivative nuttall window
minimum 4-terms discontinuous nuttall (blackman-nuttall) window
blackman-harris window
tukey window
If enabled, use fixed number of audio samples. This improves speed when filtering with large delay. Default is disabled.
Enable multichannels evaluation on gain. Default is disabled.
Enable zero phase mode by subtracting timestamp to compensate delay. Default is disabled.
Set scale used by gain. Acceptable values are:
linear frequency, linear gain
linear frequency, logarithmic (in dB) gain (default)
logarithmic (in octave scale where 20 Hz is 0) frequency, linear gain
logarithmic frequency, logarithmic gain
Set file for dumping, suitable for gnuplot.
Set scale for dumpfile. Acceptable values are same with scale option. Default is linlog.
Enable 2-channel convolution using complex FFT. This improves speed significantly. Default is disabled.
Enable minimum phase impulse response. Default is disabled.
firequalizer=gain='if(lt(f,1000), 0, -INF)'
firequalizer=gain_entry='entry(1000,0); entry(1001, -INF)'
firequalizer=gain_entry='entry(100,0); entry(400, -4); entry(1000, -6); entry(2000, 0)'
firequalizer=delay=0.1:fixed=on:zero_phase=on
firequalizer=gain='if(eq(chid,1), gain_interpolate(f), if(eq(chid,2), gain_interpolate(1e6+f), 0))' :gain_entry='entry(1000, 0); entry(1001,-INF); entry(1e6+1000,0)':multi=on
Apply a flanging effect to the audio.
The filter accepts the following options:
Set base delay in milliseconds. Range from 0 to 30. Default value is 0.
Set added sweep delay in milliseconds. Range from 0 to 10. Default value is 2.
Set percentage regeneration (delayed signal feedback). Range from -95 to 95. Default value is 0.
Set percentage of delayed signal mixed with original. Range from 0 to 100. Default value is 71.
Set sweeps per second (Hz). Range from 0.1 to 10. Default value is 0.5.
Set swept wave shape, can be triangular or sinusoidal. Default value is sinusoidal.
Set swept wave percentage-shift for multi channel. Range from 0 to 100. Default value is 25.
Set delay-line interpolation, linear or quadratic. Default is linear.
Apply Haas effect to audio.
Note that this makes most sense to apply on mono signals. With this filter applied to mono signals it give some directionality and stretches its stereo image.
The filter accepts the following options:
Set input level. By default is 1, or 0dB
Set output level. By default is 1, or 0dB.
Set gain applied to side part of signal. By default is 1.
Set kind of middle source. Can be one of the following:
Pick left channel.
Pick right channel.
Pick middle part signal of stereo image.
Pick side part signal of stereo image.
Change middle phase. By default is disabled.
Set left channel delay. By default is 2.05 milliseconds.
Set left channel balance. By default is -1.
Set left channel gain. By default is 1.
Change left phase. By default is disabled.
Set right channel delay. By defaults is 2.12 milliseconds.
Set right channel balance. By default is 1.
Set right channel gain. By default is 1.
Change right phase. By default is enabled.
Decodes High Definition Compatible Digital (HDCD) data. A 16-bit PCM stream with embedded HDCD codes is expanded into a 20-bit PCM stream.
The filter supports the Peak Extend and Low-level Gain Adjustment features of HDCD, and detects the Transient Filter flag.
ffmpeg -i HDCD16.flac -af hdcd OUT24.flac
When using the filter with wav, note the default encoding for wav is 16-bit,
so the resulting 20-bit stream will be truncated back to 16-bit. Use something
like -acodec pcm_s24le
after the filter to get 24-bit PCM output.
ffmpeg -i HDCD16.wav -af hdcd OUT16.wav ffmpeg -i HDCD16.wav -af hdcd -c:a pcm_s24le OUT24.wav
The filter accepts the following options:
Disable any automatic format conversion or resampling in the filter graph.
Process the stereo channels together. If target_gain does not match between channels, consider it invalid and use the last valid target_gain.
Set the code detect timer period in ms.
Always extend peaks above -3dBFS even if PE isn’t signaled.
Replace audio with a solid tone and adjust the amplitude to signal some specific aspect of the decoding process. The output file can be loaded in an audio editor alongside the original to aid analysis.
analyze_mode=pe:force_pe=true
can be used to see all samples above the PE level.
Modes are:
Disabled
Gain adjustment level at each sample
Samples where peak extend occurs
Samples where the code detect timer is active
Samples where the target gain does not match between channels
Apply head-related transfer functions (HRTFs) to create virtual loudspeakers around the user for binaural listening via headphones. The HRIRs are provided via additional streams, for each channel one stereo input stream is needed.
The filter accepts the following options:
Set mapping of input streams for convolution. The argument is a ’|’-separated list of channel names in order as they are given as additional stream inputs for filter. This also specify number of input streams. Number of input streams must be not less than number of channels in first stream plus one.
Set gain applied to audio. Value is in dB. Default is 0.
Set processing type. Can be time or freq. time is processing audio in time domain which is slow. freq is processing audio in frequency domain which is fast. Default is freq.
Set custom gain for LFE channels. Value is in dB. Default is 0.
Set size of frame in number of samples which will be processed at once. Default value is 1024. Allowed range is from 1024 to 96000.
Set format of hrir stream. Default value is stereo. Alternative value is multich. If value is set to stereo, number of additional streams should be greater or equal to number of input channels in first input stream. Also each additional stream should have stereo number of channels. If value is set to multich, number of additional streams should be exactly one. Also number of input channels of additional stream should be equal or greater than twice number of channels of first input stream.
ffmpeg -i input.wav -filter_complex "amovie=azi_270_ele_0_DFC.wav[sr];amovie=azi_90_ele_0_DFC.wav[sl];amovie=azi_225_ele_0_DFC.wav[br];amovie=azi_135_ele_0_DFC.wav[bl];amovie=azi_0_ele_0_DFC.wav,asplit[fc][lfe];amovie=azi_35_ele_0_DFC.wav[fl];amovie=azi_325_ele_0_DFC.wav[fr];[0:a][fl][fr][fc][lfe][bl][br][sl][sr]headphone=FL|FR|FC|LFE|BL|BR|SL|SR" output.wav
ffmpeg -i input.wav -filter_complex "amovie=minp.wav[hrirs];[0:a][hrirs]headphone=map=FL|FR|FC|LFE|BL|BR|SL|SR:hrir=multich" output.wav
Apply a high-pass filter with 3dB point frequency. The filter can be either single-pole, or double-pole (the default). The filter roll off at 6dB per pole per octave (20dB per pole per decade).
The filter accepts the following options:
Set frequency in Hz. Default is 3000.
Set number of poles. Default is 2.
Set method to specify band-width of filter.
Hz
Q-Factor
octave
slope
kHz
Specify the band-width of a filter in width_type units. Applies only to double-pole filter. The default is 0.707q and gives a Butterworth response.
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Specify which channels to filter, by default all available are filtered.
Normalize biquad coefficients, by default is disabled. Enabling it will normalize magnitude response at DC to 0dB.
Set transform type of IIR filter.
Set precision of filtering.
Pick automatic sample format depending on surround filters.
Always use signed 16-bit.
Always use signed 32-bit.
Always use float 32-bit.
Always use float 64-bit.
Set block size used for reverse IIR processing. If this value is set to high enough value (higher than impulse response length truncated when reaches near zero values) filtering will become linear phase otherwise if not big enough it will just produce nasty artifacts.
Note that filter delay will be exactly this many samples when set to non-zero value.
This filter supports the following commands:
Change highpass frequency. Syntax for the command is : "frequency"
Change highpass width_type. Syntax for the command is : "width_type"
Change highpass width. Syntax for the command is : "width"
Change highpass mix. Syntax for the command is : "mix"
Join multiple input streams into one multi-channel stream.
It accepts the following parameters:
The number of input streams. It defaults to 2.
The desired output channel layout. It defaults to stereo.
Map channels from inputs to output. The argument is a ’|’-separated list of
mappings, each in the input_idx.in_channel-out_channel
form. input_idx is the 0-based index of the input stream. in_channel
can be either the name of the input channel (e.g. FL for front left) or its
index in the specified input stream. out_channel is the name of the output
channel.
The filter will attempt to guess the mappings when they are not specified explicitly. It does so by first trying to find an unused matching input channel and if that fails it picks the first unused input channel.
Join 3 inputs (with properly set channel layouts):
ffmpeg -i INPUT1 -i INPUT2 -i INPUT3 -filter_complex join=inputs=3 OUTPUT
Build a 5.1 output from 6 single-channel streams:
ffmpeg -i fl -i fr -i fc -i sl -i sr -i lfe -filter_complex 'join=inputs=6:channel_layout=5.1:map=0.0-FL|1.0-FR|2.0-FC|3.0-SL|4.0-SR|5.0-LFE' out
Load a LADSPA (Linux Audio Developer’s Simple Plugin API) plugin.
To enable compilation of this filter you need to configure FFmpeg with
--enable-ladspa
.
Specifies the name of LADSPA plugin library to load. If the environment
variable LADSPA_PATH
is defined, the LADSPA plugin is searched in
each one of the directories specified by the colon separated list in
LADSPA_PATH
, otherwise in the standard LADSPA paths, which are in
this order: HOME/.ladspa/lib/, /usr/local/lib/ladspa/,
/usr/lib/ladspa/.
Specifies the plugin within the library. Some libraries contain only one plugin, but others contain many of them. If this is not set filter will list all available plugins within the specified library.
Set the ’|’ separated list of controls which are zero or more floating point
values that determine the behavior of the loaded plugin (for example delay,
threshold or gain).
Controls need to be defined using the following syntax:
c0=value0|c1=value1|c2=value2|..., where
valuei is the value set on the i-th control.
Alternatively they can be also defined using the following syntax:
value0|value1|value2|..., where
valuei is the value set on the i-th control.
If controls is set to help
, all available controls and
their valid ranges are printed.
Specify the sample rate, default to 44100. Only used if plugin have zero inputs.
Set the number of samples per channel per each output frame, default is 1024. Only used if plugin have zero inputs.
Set the minimum duration of the sourced audio. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax. Note that the resulting duration may be greater than the specified duration, as the generated audio is always cut at the end of a complete frame. If not specified, or the expressed duration is negative, the audio is supposed to be generated forever. Only used if plugin have zero inputs.
Enable latency compensation, by default is disabled. Only used if plugin have inputs.
ladspa=file=amp
vcf_notch
plugin from VCF
library:
ladspa=f=vcf:p=vcf_notch:c=help
Computer Music Toolkit
(CMT)
plugin library:
ladspa=file=cmt:plugin=lofi:controls=c0=22|c1=12|c2=12
ladspa=file=tap_reverb:tap_reverb
ladspa=file=cmt:noise_source_white:c=c0=.2
C* Click - Metronome
from the
C* Audio Plugin Suite
(CAPS) library:
ladspa=file=caps:Click:c=c1=20'
C* Eq10X2 - Stereo 10-band equaliser
effect:
ladspa=caps:Eq10X2:c=c0=-48|c9=-24|c3=12|c4=2
SWH Plugins
collection:
ladspa=fast_lookahead_limiter_1913:fastLookaheadLimiter:20|0|2
SWH Plugins
collection:
ladspa=mbeq_1197:mbeq:-24|-24|-24|0|0|0|0|0|0|0|0|0|0|0|0
Narrower
from the C* Audio Plugin Suite
(CAPS) library:
ladspa=caps:Narrower
C* Audio Plugin Suite
(CAPS) library:
ladspa=caps:White:.2
C* Audio Plugin Suite
(CAPS) library:
ladspa=caps:Fractal:c=c1=1
VLevel
plugin:
ladspa=vlevel-ladspa:vlevel_mono
This filter supports the following commands:
Modify the N-th control value.
If the specified value is not valid, it is ignored and prior one is kept.
EBU R128 loudness normalization. Includes both dynamic and linear normalization modes.
Support for both single pass (livestreams, files) and double pass (files) modes.
This algorithm can target IL, LRA, and maximum true peak. In dynamic mode, to accurately
detect true peaks, the audio stream will be upsampled to 192 kHz.
Use the -ar
option or aresample
filter to explicitly set an output sample rate.
The filter accepts the following options:
Set integrated loudness target. Range is -70.0 - -5.0. Default value is -24.0.
Set loudness range target. Range is 1.0 - 50.0. Default value is 7.0.
Set maximum true peak. Range is -9.0 - +0.0. Default value is -2.0.
Measured IL of input file. Range is -99.0 - +0.0.
Measured LRA of input file. Range is 0.0 - 99.0.
Measured true peak of input file. Range is -99.0 - +99.0.
Measured threshold of input file. Range is -99.0 - +0.0.
Set offset gain. Gain is applied before the true-peak limiter. Range is -99.0 - +99.0. Default is +0.0.
Normalize by linearly scaling the source audio.
measured_I
, measured_LRA
, measured_TP
,
and measured_thresh
must all be specified. Target LRA shouldn’t
be lower than source LRA and the change in integrated loudness shouldn’t
result in a true peak which exceeds the target TP. If any of these
conditions aren’t met, normalization mode will revert to dynamic.
Options are true
or false
. Default is true
.
Treat mono input files as "dual-mono". If a mono file is intended for playback
on a stereo system, its EBU R128 measurement will be perceptually incorrect.
If set to true
, this option will compensate for this effect.
Multi-channel input files are not affected by this option.
Options are true or false. Default is false.
Set print format for stats. Options are summary, json, or none. Default value is none.
Apply a low-pass filter with 3dB point frequency. The filter can be either single-pole or double-pole (the default). The filter roll off at 6dB per pole per octave (20dB per pole per decade).
The filter accepts the following options:
Set frequency in Hz. Default is 500.
Set number of poles. Default is 2.
Set method to specify band-width of filter.
Hz
Q-Factor
octave
slope
kHz
Specify the band-width of a filter in width_type units. Applies only to double-pole filter. The default is 0.707q and gives a Butterworth response.
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Specify which channels to filter, by default all available are filtered.
Normalize biquad coefficients, by default is disabled. Enabling it will normalize magnitude response at DC to 0dB.
Set transform type of IIR filter.
Set precision of filtering.
Pick automatic sample format depending on surround filters.
Always use signed 16-bit.
Always use signed 32-bit.
Always use float 32-bit.
Always use float 64-bit.
Set block size used for reverse IIR processing. If this value is set to high enough value (higher than impulse response length truncated when reaches near zero values) filtering will become linear phase otherwise if not big enough it will just produce nasty artifacts.
Note that filter delay will be exactly this many samples when set to non-zero value.
lowpass=c=LFE
This filter supports the following commands:
Change lowpass frequency. Syntax for the command is : "frequency"
Change lowpass width_type. Syntax for the command is : "width_type"
Change lowpass width. Syntax for the command is : "width"
Change lowpass mix. Syntax for the command is : "mix"
Load a LV2 (LADSPA Version 2) plugin.
To enable compilation of this filter you need to configure FFmpeg with
--enable-lv2
.
Specifies the plugin URI. You may need to escape ’:’.
Set the ’|’ separated list of controls which are zero or more floating point
values that determine the behavior of the loaded plugin (for example delay,
threshold or gain).
If controls is set to help
, all available controls and
their valid ranges are printed.
Specify the sample rate, default to 44100. Only used if plugin have zero inputs.
Set the number of samples per channel per each output frame, default is 1024. Only used if plugin have zero inputs.
Set the minimum duration of the sourced audio. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax. Note that the resulting duration may be greater than the specified duration, as the generated audio is always cut at the end of a complete frame. If not specified, or the expressed duration is negative, the audio is supposed to be generated forever. Only used if plugin have zero inputs.
lv2=p=http\\\\://calf.sourceforge.net/plugins/BassEnhancer:c=amount=2
lv2=p=http\\\\://calf.sourceforge.net/plugins/Vinyl:c=drone=0.2|aging=0.5
lv2=p=http\\\\://www.openavproductions.com/artyfx#bitta:c=crush=0.3
This filter supports all options that are exported by plugin as commands.
Multiband Compress or expand the audio’s dynamic range.
The input audio is divided into bands using 4th order Linkwitz-Riley IIRs. This is akin to the crossover of a loudspeaker, and results in flat frequency response when absent compander action.
It accepts the following parameters:
This option syntax is: attack,decay,[attack,decay..] soft-knee points crossover_frequency [delay [initial_volume [gain]]] | attack,decay ... For explanation of each item refer to compand filter documentation.
Mix channels with specific gain levels. The filter accepts the output channel layout followed by a set of channels definitions.
This filter is also designed to efficiently remap the channels of an audio stream.
The filter accepts parameters of the form: "l|outdef|outdef|..."
output channel layout or number of channels
output channel specification, of the form: "out_name=[gain*]in_name[(+-)[gain*]in_name...]"
output channel to define, either a channel name (FL, FR, etc.) or a channel number (c0, c1, etc.)
multiplicative coefficient for the channel, 1 leaving the volume unchanged
input channel to use, see out_name for details; it is not possible to mix named and numbered input channels
If the ‘=’ in a channel specification is replaced by ‘<’, then the gains for that specification will be renormalized so that the total is 1, thus avoiding clipping noise.
For example, if you want to down-mix from stereo to mono, but with a bigger factor for the left channel:
pan=1c|c0=0.9*c0+0.1*c1
A customized down-mix to stereo that works automatically for 3-, 4-, 5- and 7-channels surround:
pan=stereo| FL < FL + 0.5*FC + 0.6*BL + 0.6*SL | FR < FR + 0.5*FC + 0.6*BR + 0.6*SR
Note that ffmpeg
integrates a default down-mix (and up-mix) system
that should be preferred (see "-ac" option) unless you have very specific
needs.
The channel remapping will be effective if, and only if:
If all these conditions are satisfied, the filter will notify the user ("Pure channel mapping detected"), and use an optimized and lossless method to do the remapping.
For example, if you have a 5.1 source and want a stereo audio stream by dropping the extra channels:
pan="stereo| c0=FL | c1=FR"
Given the same source, you can also switch front left and front right channels and keep the input channel layout:
pan="5.1| c0=c1 | c1=c0 | c2=c2 | c3=c3 | c4=c4 | c5=c5"
If the input is a stereo audio stream, you can mute the front left channel (and still keep the stereo channel layout) with:
pan="stereo|c1=c1"
Still with a stereo audio stream input, you can copy the right channel in both front left and right:
pan="stereo| c0=FR | c1=FR"
ReplayGain scanner filter. This filter takes an audio stream as an input and
outputs it unchanged.
At end of filtering it displays track_gain
and track_peak
.
The filter accepts the following exported read-only options:
Exported track gain in dB at end of stream.
Exported track peak at end of stream.
Convert the audio sample format, sample rate and channel layout. It is not meant to be used directly.
Apply time-stretching and pitch-shifting with librubberband.
To enable compilation of this filter, you need to configure FFmpeg with
--enable-librubberband
.
The filter accepts the following options:
Set tempo scale factor.
Set pitch scale factor.
Set transients detector. Possible values are:
Set detector. Possible values are:
Set phase. Possible values are:
Set processing window size. Possible values are:
Set smoothing. Possible values are:
Enable formant preservation when shift pitching. Possible values are:
Set pitch quality. Possible values are:
Set channels. Possible values are:
This filter supports the following commands:
Change filter tempo scale factor. Syntax for the command is : "tempo"
Change filter pitch scale factor. Syntax for the command is : "pitch"
This filter acts like normal compressor but has the ability to compress detected signal using second input signal. It needs two input streams and returns one output stream. First input stream will be processed depending on second stream signal. The filtered signal then can be filtered with other filters in later stages of processing. See pan and amerge filter.
The filter accepts the following options:
Set input gain. Default is 1. Range is between 0.015625 and 64.
Set mode of compressor operation. Can be upward
or downward
.
Default is downward
.
If a signal of second stream raises above this level it will affect the gain reduction of first stream. By default is 0.125. Range is between 0.00097563 and 1.
Set a ratio about which the signal is reduced. 1:2 means that if the level raised 4dB above the threshold, it will be only 2dB above after the reduction. Default is 2. Range is between 1 and 20.
Amount of milliseconds the signal has to rise above the threshold before gain reduction starts. Default is 20. Range is between 0.01 and 2000.
Amount of milliseconds the signal has to fall below the threshold before reduction is decreased again. Default is 250. Range is between 0.01 and 9000.
Set the amount by how much signal will be amplified after processing. Default is 1. Range is from 1 to 64.
Curve the sharp knee around the threshold to enter gain reduction more softly. Default is 2.82843. Range is between 1 and 8.
Choose if the average
level between all channels of side-chain stream
or the louder(maximum
) channel of side-chain stream affects the
reduction. Default is average
.
Should the exact signal be taken in case of peak
or an RMS one in case
of rms
. Default is rms
which is mainly smoother.
Set sidechain gain. Default is 1. Range is between 0.015625 and 64.
How much to use compressed signal in output. Default is 1. Range is between 0 and 1.
This filter supports the all above options as commands.
ffmpeg -i main.flac -i sidechain.flac -filter_complex "[1:a]asplit=2[sc][mix];[0:a][sc]sidechaincompress[compr];[compr][mix]amerge"
A sidechain gate acts like a normal (wideband) gate but has the ability to filter the detected signal before sending it to the gain reduction stage. Normally a gate uses the full range signal to detect a level above the threshold. For example: If you cut all lower frequencies from your sidechain signal the gate will decrease the volume of your track only if not enough highs appear. With this technique you are able to reduce the resonation of a natural drum or remove "rumbling" of muted strokes from a heavily distorted guitar. It needs two input streams and returns one output stream. First input stream will be processed depending on second stream signal.
The filter accepts the following options:
Set input level before filtering. Default is 1. Allowed range is from 0.015625 to 64.
Set the mode of operation. Can be upward
or downward
.
Default is downward
. If set to upward
mode, higher parts of signal
will be amplified, expanding dynamic range in upward direction.
Otherwise, in case of downward
lower parts of signal will be reduced.
Set the level of gain reduction when the signal is below the threshold. Default is 0.06125. Allowed range is from 0 to 1. Setting this to 0 disables reduction and then filter behaves like expander.
If a signal rises above this level the gain reduction is released. Default is 0.125. Allowed range is from 0 to 1.
Set a ratio about which the signal is reduced. Default is 2. Allowed range is from 1 to 9000.
Amount of milliseconds the signal has to rise above the threshold before gain reduction stops. Default is 20 milliseconds. Allowed range is from 0.01 to 9000.
Amount of milliseconds the signal has to fall below the threshold before the reduction is increased again. Default is 250 milliseconds. Allowed range is from 0.01 to 9000.
Set amount of amplification of signal after processing. Default is 1. Allowed range is from 1 to 64.
Curve the sharp knee around the threshold to enter gain reduction more softly. Default is 2.828427125. Allowed range is from 1 to 8.
Choose if exact signal should be taken for detection or an RMS like one. Default is rms. Can be peak or rms.
Choose if the average level between all channels or the louder channel affects the reduction. Default is average. Can be average or maximum.
Set sidechain gain. Default is 1. Range is from 0.015625 to 64.
This filter supports the all above options as commands.
Detect silence in an audio stream.
This filter logs a message when it detects that the input audio volume is less or equal to a noise tolerance value for a duration greater or equal to the minimum detected noise duration.
The printed times and duration are expressed in seconds. The
lavfi.silence_start
or lavfi.silence_start.X
metadata key
is set on the first frame whose timestamp equals or exceeds the detection
duration and it contains the timestamp of the first frame of the silence.
The lavfi.silence_duration
or lavfi.silence_duration.X
and lavfi.silence_end
or lavfi.silence_end.X
metadata
keys are set on the first frame after the silence. If mono is
enabled, and each channel is evaluated separately, the .X
suffixed keys are used, and X
corresponds to the channel number.
The filter accepts the following options:
Set noise tolerance. Can be specified in dB (in case "dB" is appended to the specified value) or amplitude ratio. Default is -60dB, or 0.001.
Set silence duration until notification (default is 2 seconds). See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax.
Process each channel separately, instead of combined. By default is disabled.
silencedetect=n=-50dB:d=5
ffmpeg
to detect silence with 0.0001 noise
tolerance in silence.mp3:
ffmpeg -i silence.mp3 -af silencedetect=noise=0.0001 -f null -
Remove silence from the beginning, middle or end of the audio.
The filter accepts the following options:
This value is used to indicate if audio should be trimmed at beginning of
the audio. A value of zero indicates no silence should be trimmed from the
beginning. When specifying a non-zero value, it trims audio up until it
finds non-silence. Normally, when trimming silence from beginning of audio
the start_periods will be 1
but it can be increased to higher
values to trim all audio up to specific count of non-silence periods.
Default value is 0
.
Specify the amount of time that non-silence must be detected before it stops
trimming audio. By increasing the duration, bursts of noises can be treated
as silence and trimmed off. Default value is 0
.
This indicates what sample value should be treated as silence. For digital
audio, a value of 0
may be fine but for audio recorded from analog,
you may wish to increase the value to account for background noise.
Can be specified in dB (in case "dB" is appended to the specified value)
or amplitude ratio. Default value is 0
.
Specify max duration of silence at beginning that will be kept after trimming. Default is 0, which is equal to trimming all samples detected as silence.
Specify mode of detection of silence end at start of multi-channel audio. Can be any or all. Default is any. With any, any sample from any channel that is detected as non-silence will trigger end of silence trimming at start of audio stream. With all, only if every sample from every channel is detected as non-silence will trigger end of silence trimming at start of audio stream, limited usage.
Set the count for trimming silence from the end of audio. When specifying a
positive value, it trims audio after it finds specified silence period.
To remove silence from the middle of a file, specify a stop_periods
that is negative. This value is then treated as a positive value and is
used to indicate the effect should restart processing as specified by
stop_periods, making it suitable for removing periods of silence
in the middle of the audio.
Default value is 0
.
Specify a duration of silence that must exist before audio is not copied any
more. By specifying a higher duration, silence that is wanted can be left in
the audio.
Default value is 0
.
This is the same as start_threshold but for trimming silence from
the end of audio.
Can be specified in dB (in case "dB" is appended to the specified value)
or amplitude ratio. Default value is 0
.
Specify max duration of silence at end that will be kept after trimming. Default is 0, which is equal to trimming all samples detected as silence.
Specify mode of detection of silence start after start of multi-channel audio. Can be any or all. Default is all. With any, any sample from any channel that is detected as silence will trigger start of silence trimming after start of audio stream, limited usage. With all, only if every sample from every channel is detected as silence will trigger start of silence trimming after start of audio stream.
Set how is silence detected.
Mean of absolute values of samples in moving window.
Root squared mean of absolute values of samples in moving window.
Maximum of absolute values of samples in moving window.
Median of absolute values of samples in moving window.
Absolute of max peak to min peak difference of samples in moving window.
Standard deviation of values of samples in moving window.
Default value is rms
.
Set duration in number of seconds used to calculate size of window in number
of samples for detecting silence. Using 0
will effectively disable
any windowing and use only single sample per channel for silence detection.
In that case it may be needed to also set start_silence and/or
stop_silence to nonzero values with also start_duration and/or
stop_duration to nonzero values.
Default value is 0.02
. Allowed range is from 0
to 10
.
Set processing mode of every audio frame output timestamp.
Full timestamps rewrite, keep only the start time for the first output frame.
Non-dropped frames are left with same timestamp as input audio frame.
Defaults value is write
.
silenceremove=start_periods=1:start_duration=5:start_threshold=0.02
silenceremove=stop_periods=-1:stop_duration=1:stop_threshold=-90dB
silenceremove=window=0:detection=peak:stop_mode=all:start_mode=all:stop_periods=-1:stop_threshold=0
silenceremove=stop_periods=-2:stop_duration=1:stop_threshold=-90dB
silenceremove=stop_periods=-2:stop_duration=1:stop_threshold=-90dB:stop_silence=0.5
silenceremove=stop_periods=-2:stop_duration=1:stop_threshold=-90dB:stop_silence=0.5:start_periods=1:start_duration=1:start_silence=1.5:stop_threshold=-90dB
This filter supports some above options as commands.
SOFAlizer uses head-related transfer functions (HRTFs) to create virtual loudspeakers around the user for binaural listening via headphones (audio formats up to 9 channels supported). The HRTFs are stored in SOFA files (see http://www.sofacoustics.org/ for a database). SOFAlizer is developed at the Acoustics Research Institute (ARI) of the Austrian Academy of Sciences.
To enable compilation of this filter you need to configure FFmpeg with
--enable-libmysofa
.
The filter accepts the following options:
Set the SOFA file used for rendering.
Set gain applied to audio. Value is in dB. Default is 0.
Set rotation of virtual loudspeakers in deg. Default is 0.
Set elevation of virtual speakers in deg. Default is 0.
Set distance in meters between loudspeakers and the listener with near-field HRTFs. Default is 1.
Set processing type. Can be time or freq. time is processing audio in time domain which is slow. freq is processing audio in frequency domain which is fast. Default is freq.
Set custom positions of virtual loudspeakers. Syntax for this option is: <CH> <AZIM> <ELEV>[|<CH> <AZIM> <ELEV>|...]. Each virtual loudspeaker is described with short channel name following with azimuth and elevation in degrees. Each virtual loudspeaker description is separated by ’|’. For example to override front left and front right channel positions use: ’speakers=FL 45 15|FR 345 15’. Descriptions with unrecognised channel names are ignored.
Set custom gain for LFE channels. Value is in dB. Default is 0.
Set custom frame size in number of samples. Default is 1024. Allowed range is from 1024 to 96000. Only used if option ‘type’ is set to freq.
Should all IRs be normalized upon importing SOFA file. By default is enabled.
Should nearest IRs be interpolated with neighbor IRs if exact position does not match. By default is disabled.
Minphase all IRs upon loading of SOFA file. By default is disabled.
Set neighbor search angle step. Only used if option interpolate is enabled.
Set neighbor search radius step. Only used if option interpolate is enabled.
sofalizer=sofa=/path/to/ClubFritz6.sofa:type=freq:radius=1
sofalizer=sofa=/path/to/ClubFritz12.sofa:type=freq:radius=2:rotation=5
"sofalizer=sofa=/path/to/ClubFritz6.sofa:type=freq:radius=2:speakers=FL 45|FR 315|BL 135|BR 225:gain=28"
Speech Normalizer.
This filter expands or compresses each half-cycle of audio samples (local set of samples all above or all below zero and between two nearest zero crossings) depending on threshold value, so audio reaches target peak value under conditions controlled by below options.
The filter accepts the following options:
Set the expansion target peak value. This specifies the highest allowed absolute amplitude level for the normalized audio input. Default value is 0.95. Allowed range is from 0.0 to 1.0.
Set the maximum expansion factor. Allowed range is from 1.0 to 50.0. Default value is 2.0. This option controls maximum local half-cycle of samples expansion. The maximum expansion would be such that local peak value reaches target peak value but never to surpass it and that ratio between new and previous peak value does not surpass this option value.
Set the maximum compression factor. Allowed range is from 1.0 to 50.0. Default value is 2.0. This option controls maximum local half-cycle of samples compression. This option is used only if threshold option is set to value greater than 0.0, then in such cases when local peak is lower or same as value set by threshold all samples belonging to that peak’s half-cycle will be compressed by current compression factor.
Set the threshold value. Default value is 0.0. Allowed range is from 0.0 to 1.0. This option specifies which half-cycles of samples will be compressed and which will be expanded. Any half-cycle samples with their local peak value below or same as this option value will be compressed by current compression factor, otherwise, if greater than threshold value they will be expanded with expansion factor so that it could reach peak target value but never surpass it.
Set the expansion raising amount per each half-cycle of samples. Default value is 0.001. Allowed range is from 0.0 to 1.0. This controls how fast expansion factor is raised per each new half-cycle until it reaches expansion value. Setting this options too high may lead to distortions.
Set the compression raising amount per each half-cycle of samples. Default value is 0.001. Allowed range is from 0.0 to 1.0. This controls how fast compression factor is raised per each new half-cycle until it reaches compression value.
Specify which channels to filter, by default all available channels are filtered.
Enable inverted filtering, by default is disabled. This inverts interpretation of threshold option. When enabled any half-cycle of samples with their local peak value below or same as threshold option will be expanded otherwise it will be compressed.
Link channels when calculating gain applied to each filtered channel sample, by default is disabled. When disabled each filtered channel gain calculation is independent, otherwise when this option is enabled the minimum of all possible gains for each filtered channel is used.
Set the expansion target RMS value. This specifies the highest allowed RMS level for the normalized audio input. Default value is 0.0, thus disabled. Allowed range is from 0.0 to 1.0.
This filter supports the all above options as commands.
speechnorm=e=3:r=0.00001:l=1
speechnorm=e=6.25:r=0.00001:l=1
speechnorm=e=12.5:r=0.0001:l=1
speechnorm=e=25:r=0.0001:l=1
speechnorm=e=50:r=0.0001:l=1
This filter has some handy utilities to manage stereo signals, for converting M/S stereo recordings to L/R signal while having control over the parameters or spreading the stereo image of master track.
The filter accepts the following options:
Set input level before filtering for both channels. Defaults is 1. Allowed range is from 0.015625 to 64.
Set output level after filtering for both channels. Defaults is 1. Allowed range is from 0.015625 to 64.
Set input balance between both channels. Default is 0. Allowed range is from -1 to 1.
Set output balance between both channels. Default is 0. Allowed range is from -1 to 1.
Enable softclipping. Results in analog distortion instead of harsh digital 0dB clipping. Disabled by default.
Mute the left channel. Disabled by default.
Mute the right channel. Disabled by default.
Change the phase of the left channel. Disabled by default.
Change the phase of the right channel. Disabled by default.
Set stereo mode. Available values are:
Left/Right to Left/Right, this is default.
Left/Right to Mid/Side.
Mid/Side to Left/Right.
Left/Right to Left/Left.
Left/Right to Right/Right.
Left/Right to Left + Right.
Left/Right to Right/Left.
Mid/Side to Left/Left.
Mid/Side to Right/Right.
Mid/Side to Right/Left.
Left/Right to Left - Right.
Set level of side signal. Default is 1. Allowed range is from 0.015625 to 64.
Set balance of side signal. Default is 0. Allowed range is from -1 to 1.
Set level of the middle signal. Default is 1. Allowed range is from 0.015625 to 64.
Set middle signal pan. Default is 0. Allowed range is from -1 to 1.
Set stereo base between mono and inversed channels. Default is 0. Allowed range is from -1 to 1.
Set delay in milliseconds how much to delay left from right channel and vice versa. Default is 0. Allowed range is from -20 to 20.
Set S/C level. Default is 1. Allowed range is from 1 to 100.
Set the stereo phase in degrees. Default is 0. Allowed range is from 0 to 360.
Set balance mode for balance_in/balance_out option.
Can be one of the following:
Classic balance mode. Attenuate one channel at time. Gain is raised up to 1.
Similar as classic mode above but gain is raised up to 2.
Equal power distribution, from -6dB to +6dB range.
This filter supports the all above options as commands.
stereotools=mlev=0.015625
"stereotools=mode=ms>lr"
This filter enhance the stereo effect by suppressing signal common to both channels and by delaying the signal of left into right and vice versa, thereby widening the stereo effect.
The filter accepts the following options:
Time in milliseconds of the delay of left signal into right and vice versa. Default is 20 milliseconds.
Amount of gain in delayed signal into right and vice versa. Gives a delay effect of left signal in right output and vice versa which gives widening effect. Default is 0.3.
Cross feed of left into right with inverted phase. This helps in suppressing the mono. If the value is 1 it will cancel all the signal common to both channels. Default is 0.3.
Set level of input signal of original channel. Default is 0.8.
This filter supports the all above options except delay
as commands.
Apply 18 band equalizer.
The filter accepts the following options:
Set 65Hz band gain.
Set 92Hz band gain.
Set 131Hz band gain.
Set 185Hz band gain.
Set 262Hz band gain.
Set 370Hz band gain.
Set 523Hz band gain.
Set 740Hz band gain.
Set 1047Hz band gain.
Set 1480Hz band gain.
Set 2093Hz band gain.
Set 2960Hz band gain.
Set 4186Hz band gain.
Set 5920Hz band gain.
Set 8372Hz band gain.
Set 11840Hz band gain.
Set 16744Hz band gain.
Set 20000Hz band gain.
Apply audio surround upmix filter.
This filter allows to produce multichannel output from audio stream.
The filter accepts the following options:
Set output channel layout. By default, this is 5.1.
See (ffmpeg-utils)the Channel Layout section in the ffmpeg-utils(1) manual for the required syntax.
Set input channel layout. By default, this is stereo.
See (ffmpeg-utils)the Channel Layout section in the ffmpeg-utils(1) manual for the required syntax.
Set input volume level. By default, this is 1.
Set output volume level. By default, this is 1.
Enable LFE channel output if output channel layout has it. By default, this is enabled.
Set LFE low cut off frequency. By default, this is 128 Hz.
Set LFE high cut off frequency. By default, this is 256 Hz.
Set LFE mode, can be add or sub. Default is add. In add mode, LFE channel is created from input audio and added to output. In sub mode, LFE channel is created from input audio and added to output but also all non-LFE output channels are subtracted with output LFE channel.
Set temporal smoothness strength, used to gradually change factors when transforming stereo sound in time. Allowed range is from 0.0 to 1.0. Useful to improve output quality with focus option values greater than 0.0. Default is 0.0. Only values inside this range and without edges are effective.
Set angle of stereo surround transform, Allowed range is from 0 to 360. Default is 90.
Set focus of stereo surround transform, Allowed range is from -1 to 1. Default is 0.
Set front center input volume. By default, this is 1.
Set front center output volume. By default, this is 1.
Set front left input volume. By default, this is 1.
Set front left output volume. By default, this is 1.
Set front right input volume. By default, this is 1.
Set front right output volume. By default, this is 1.
Set side left input volume. By default, this is 1.
Set side left output volume. By default, this is 1.
Set side right input volume. By default, this is 1.
Set side right output volume. By default, this is 1.
Set back left input volume. By default, this is 1.
Set back left output volume. By default, this is 1.
Set back right input volume. By default, this is 1.
Set back right output volume. By default, this is 1.
Set back center input volume. By default, this is 1.
Set back center output volume. By default, this is 1.
Set LFE input volume. By default, this is 1.
Set LFE output volume. By default, this is 1.
Set spread usage of stereo image across X axis for all channels. Allowed range is from -1 to 15. By default this value is negative -1, and thus unused.
Set spread usage of stereo image across Y axis for all channels. Allowed range is from -1 to 15. By default this value is negative -1, and thus unused.
Set spread usage of stereo image across X axis for each channel. Allowed range is from 0.06 to 15. By default this value is 0.5.
Set spread usage of stereo image across Y axis for each channel. Allowed range is from 0.06 to 15. By default this value is 0.5.
Set window size. Allowed range is from 1024 to 65536. Default size is 4096.
Set window function.
It accepts the following values:
Default is hann
.
Set window overlap. If set to 1, the recommended overlap for selected
window function will be picked. Default is 0.5
.
Boost or cut the lower frequencies and cut or boost higher frequencies of the audio using a two-pole shelving filter with a response similar to that of a standard hi-fi’s tone-controls. This is also known as shelving equalisation (EQ).
The filter accepts the following options:
Give the gain at 0 Hz. Its useful range is about -20 (for a large cut) to +20 (for a large boost). Beware of clipping when using a positive gain.
Set the filter’s central frequency and so can be used
to extend or reduce the frequency range to be boosted or cut.
The default value is 3000
Hz.
Set method to specify band-width of filter.
Hz
Q-Factor
octave
slope
kHz
Determine how steep is the filter’s shelf transition.
Set number of poles. Default is 2.
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Specify which channels to filter, by default all available are filtered.
Normalize biquad coefficients, by default is disabled. Enabling it will normalize magnitude response at DC to 0dB.
Set transform type of IIR filter.
Set precision of filtering.
Pick automatic sample format depending on surround filters.
Always use signed 16-bit.
Always use signed 32-bit.
Always use float 32-bit.
Always use float 64-bit.
Set block size used for reverse IIR processing. If this value is set to high enough value (higher than impulse response length truncated when reaches near zero values) filtering will become linear phase otherwise if not big enough it will just produce nasty artifacts.
Note that filter delay will be exactly this many samples when set to non-zero value.
This filter supports some options as commands.
Boost or cut treble (upper) frequencies of the audio using a two-pole shelving filter with a response similar to that of a standard hi-fi’s tone-controls. This is also known as shelving equalisation (EQ).
The filter accepts the following options:
Give the gain at whichever is the lower of ~22 kHz and the Nyquist frequency. Its useful range is about -20 (for a large cut) to +20 (for a large boost). Beware of clipping when using a positive gain.
Set the filter’s central frequency and so can be used
to extend or reduce the frequency range to be boosted or cut.
The default value is 3000
Hz.
Set method to specify band-width of filter.
Hz
Q-Factor
octave
slope
kHz
Determine how steep is the filter’s shelf transition.
Set number of poles. Default is 2.
How much to use filtered signal in output. Default is 1. Range is between 0 and 1.
Specify which channels to filter, by default all available are filtered.
Normalize biquad coefficients, by default is disabled. Enabling it will normalize magnitude response at DC to 0dB.
Set transform type of IIR filter.
Set precision of filtering.
Pick automatic sample format depending on surround filters.
Always use signed 16-bit.
Always use signed 32-bit.
Always use float 32-bit.
Always use float 64-bit.
Set block size used for reverse IIR processing. If this value is set to high enough value (higher than impulse response length truncated when reaches near zero values) filtering will become linear phase otherwise if not big enough it will just produce nasty artifacts.
Note that filter delay will be exactly this many samples when set to non-zero value.
This filter supports the following commands:
Change treble frequency. Syntax for the command is : "frequency"
Change treble width_type. Syntax for the command is : "width_type"
Change treble width. Syntax for the command is : "width"
Change treble gain. Syntax for the command is : "gain"
Change treble mix. Syntax for the command is : "mix"
Sinusoidal amplitude modulation.
The filter accepts the following options:
Modulation frequency in Hertz. Modulation frequencies in the subharmonic range (20 Hz or lower) will result in a tremolo effect. This filter may also be used as a ring modulator by specifying a modulation frequency higher than 20 Hz. Range is 0.1 - 20000.0. Default value is 5.0 Hz.
Depth of modulation as a percentage. Range is 0.0 - 1.0. Default value is 0.5.
Sinusoidal phase modulation.
The filter accepts the following options:
Modulation frequency in Hertz. Range is 0.1 - 20000.0. Default value is 5.0 Hz.
Depth of modulation as a percentage. Range is 0.0 - 1.0. Default value is 0.5.
Apply audio Virtual Bass filter.
This filter accepts stereo input and produce stereo with LFE (2.1) channels output. The newly produced LFE channel have enhanced virtual bass originally obtained from both stereo channels. This filter outputs front left and front right channels unchanged as available in stereo input.
The filter accepts the following options:
Set the virtual bass cutoff frequency. Default value is 250 Hz. Allowed range is from 100 to 500 Hz.
Set the virtual bass strength. Allowed range is from 0.5 to 3. Default value is 3.
Adjust the input audio volume.
It accepts the following parameters:
Set audio volume expression.
Output values are clipped to the maximum value.
The output audio volume is given by the relation:
output_volume = volume * input_volume
The default value for volume is "1.0".
This parameter represents the mathematical precision.
It determines which input sample formats will be allowed, which affects the precision of the volume scaling.
8-bit fixed-point; this limits input sample format to U8, S16, and S32.
32-bit floating-point; this limits input sample format to FLT. (default)
64-bit floating-point; this limits input sample format to DBL.
Choose the behaviour on encountering ReplayGain side data in input frames.
Remove ReplayGain side data, ignoring its contents (the default).
Ignore ReplayGain side data, but leave it in the frame.
Prefer the track gain, if present.
Prefer the album gain, if present.
Pre-amplification gain in dB to apply to the selected replaygain gain.
Default value for replaygain_preamp is 0.0.
Prevent clipping by limiting the gain applied.
Default value for replaygain_noclip is 1.
Set when the volume expression is evaluated.
It accepts the following values:
only evaluate expression once during the filter initialization, or when the ‘volume’ command is sent
evaluate expression for each incoming frame
Default value is ‘once’.
The volume expression can contain the following parameters.
frame number (starting at zero)
number of channels
number of samples consumed by the filter
number of samples in the current frame
original frame position in the file; deprecated, do not use
frame PTS
sample rate
PTS at start of stream
time at start of stream
frame time
timestamp timebase
last set volume value
Note that when eval is set to ‘once’ only the sample_rate and tb variables are available, all other variables will evaluate to NAN.
This filter supports the following commands:
Modify the volume expression. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
volume=volume=0.5 volume=volume=1/2 volume=volume=-6.0206dB
In all the above example the named key for volume can be omitted, for example like in:
volume=0.5
volume=volume=6dB:precision=fixed
volume='if(lt(t,10),1,max(1-(t-10)/5,0))':eval=frame
Detect the volume of the input video.
The filter has no parameters. It supports only 16-bit signed integer samples, so the input will be converted when needed. Statistics about the volume will be printed in the log when the input stream end is reached.
In particular it will show the mean volume (root mean square), maximum volume (on a per-sample basis), and the beginning of a histogram of the registered volume values (from the maximum value to a cumulated 1/1000 of the samples).
All volumes are in decibels relative to the maximum PCM value.
Here is an excerpt of the output:
[Parsed_volumedetect_0 0xa23120] mean_volume: -27 dB [Parsed_volumedetect_0 0xa23120] max_volume: -4 dB [Parsed_volumedetect_0 0xa23120] histogram_4db: 6 [Parsed_volumedetect_0 0xa23120] histogram_5db: 62 [Parsed_volumedetect_0 0xa23120] histogram_6db: 286 [Parsed_volumedetect_0 0xa23120] histogram_7db: 1042 [Parsed_volumedetect_0 0xa23120] histogram_8db: 2551 [Parsed_volumedetect_0 0xa23120] histogram_9db: 4609 [Parsed_volumedetect_0 0xa23120] histogram_10db: 8409
It means that:
In other words, raising the volume by +4 dB does not cause any clipping, raising it by +5 dB causes clipping for 6 samples, etc.
Below is a description of the currently available audio sources.
Buffer audio frames, and make them available to the filter chain.
This source is mainly intended for a programmatic use, in particular through the interface defined in libavfilter/buffersrc.h.
It accepts the following parameters:
The timebase which will be used for timestamps of submitted frames. It must be either a floating-point number or in numerator/denominator form.
The sample rate of the incoming audio buffers.
The sample format of the incoming audio buffers. Either a sample format name or its corresponding integer representation from the enum AVSampleFormat in libavutil/samplefmt.h
The channel layout of the incoming audio buffers. Either a channel layout name from channel_layout_map in libavutil/channel_layout.c or its corresponding integer representation from the AV_CH_LAYOUT_* macros in libavutil/channel_layout.h
The number of channels of the incoming audio buffers. If both channels and channel_layout are specified, then they must be consistent.
abuffer=sample_rate=44100:sample_fmt=s16p:channel_layout=stereo
will instruct the source to accept planar 16bit signed stereo at 44100Hz. Since the sample format with name "s16p" corresponds to the number 6 and the "stereo" channel layout corresponds to the value 0x3, this is equivalent to:
abuffer=sample_rate=44100:sample_fmt=6:channel_layout=0x3
Generate an audio signal specified by an expression.
This source accepts in input one or more expressions (one for each channel), which are evaluated and used to generate a corresponding audio signal.
This source accepts the following options:
Set the ’|’-separated expressions list for each separate channel. In case the channel_layout option is not specified, the selected channel layout depends on the number of provided expressions. Otherwise the last specified expression is applied to the remaining output channels.
Set the channel layout. The number of channels in the specified layout must be equal to the number of specified expressions.
Set the minimum duration of the sourced audio. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax. Note that the resulting duration may be greater than the specified duration, as the generated audio is always cut at the end of a complete frame.
If not specified, or the expressed duration is negative, the audio is supposed to be generated forever.
Set the number of samples per channel per each output frame, default to 1024.
Specify the sample rate, default to 44100.
Each expression in exprs can contain the following constants:
number of the evaluated sample, starting from 0
time of the evaluated sample expressed in seconds, starting from 0
sample rate
aevalsrc=0
aevalsrc="sin(440*2*PI*t):s=8000"
aevalsrc="sin(420*2*PI*t)|cos(430*2*PI*t):c=FC|BC"
aevalsrc="-2+random(0)"
aevalsrc="sin(10*2*PI*t)*sin(880*2*PI*t)"
aevalsrc="0.1*sin(2*PI*(360-2.5/2)*t) | 0.1*sin(2*PI*(360+2.5/2)*t)"
Generate a fractional delay FIR coefficients.
The resulting stream can be used with afir filter for filtering the audio signal.
The filter accepts the following options:
Set the fractional delay. Default is 0.
Set the sample rate, default is 44100.
Set the number of samples per each frame. Default is 1024.
Set the number of filter coefficients in output audio stream. Default value is 0.
Specifies the channel layout, and can be a string representing a channel layout. The default value of channel_layout is "stereo".
Generate a FIR equalizer coefficients.
The resulting stream can be used with afir filter for filtering the audio signal.
The filter accepts the following options:
Set equalizer preset.
Default preset is flat
.
Available presets are:
Set custom gains for each band. Only used if the preset option is set to custom
.
Gains are separated by white spaces and each gain is set in dBFS.
Default is 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0 0
.
Set the custom bands from where custon equalizer gains are set.
This must be in strictly increasing order. Only used if the preset option is set to custom
.
Bands are separated by white spaces and each band represent frequency in Hz.
Default is 25 40 63 100 160 250 400 630 1000 1600 2500 4000 6300 10000 16000 24000
.
Set number of filter coefficients in output audio stream.
Default value is 4096
.
Set sample rate of output audio stream, default is 44100
.
Set number of samples per each frame in output audio stream. Default is 1024
.
Set interpolation method for FIR equalizer coefficients. Can be linear
or cubic
.
Set phase type of FIR filter. Can be linear
or min
: minimum-phase.
Default is minimum-phase filter.
Generate a FIR coefficients using frequency sampling method.
The resulting stream can be used with afir filter for filtering the audio signal.
The filter accepts the following options:
Set number of filter coefficients in output audio stream. Default value is 1025.
Set frequency points from where magnitude and phase are set. This must be in non decreasing order, and first element must be 0, while last element must be 1. Elements are separated by white spaces.
Set magnitude value for every frequency point set by frequency. Number of values must be same as number of frequency points. Values are separated by white spaces.
Set phase value for every frequency point set by frequency. Number of values must be same as number of frequency points. Values are separated by white spaces.
Set sample rate, default is 44100.
Set number of samples per each frame. Default is 1024.
Set window function. Default is blackman.
The null audio source, return unprocessed audio frames. It is mainly useful as a template and to be employed in analysis / debugging tools, or as the source for filters which ignore the input data (for example the sox synth filter).
This source accepts the following options:
Specifies the channel layout, and can be either an integer or a string representing a channel layout. The default value of channel_layout is "stereo".
Check the channel_layout_map definition in libavutil/channel_layout.c for the mapping between strings and channel layout values.
Specifies the sample rate, and defaults to 44100.
Set the number of samples per requested frames.
Set the duration of the sourced audio. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax.
If not specified, or the expressed duration is negative, the audio is supposed to be generated forever.
anullsrc=r=48000:cl=4
anullsrc=r=48000:cl=mono
All the parameters need to be explicitly defined.
Synthesize a voice utterance using the libflite library.
To enable compilation of this filter you need to configure FFmpeg with
--enable-libflite
.
Note that versions of the flite library prior to 2.0 are not thread-safe.
The filter accepts the following options:
If set to 1, list the names of the available voices and exit immediately. Default value is 0.
Set the maximum number of samples per frame. Default value is 512.
Set the filename containing the text to speak.
Set the text to speak.
Set the voice to use for the speech synthesis. Default value is
kal
. See also the list_voices option.
flite=textfile=speech.txt
slt
voice:
flite=text='So fare thee well, poor devil of a Sub-Sub, whose commentator I am':voice=slt
ffmpeg -f lavfi -i flite=text='So fare thee well, poor devil of a Sub-Sub, whose commentator I am':voice=slt
flite
and
the lavfi
device:
ffplay -f lavfi flite=text='No more be grieved for which that thou hast done.'
For more information about libflite, check: http://www.festvox.org/flite/
Generate a noise audio signal.
The filter accepts the following options:
Specify the sample rate. Default value is 48000 Hz.
Specify the amplitude (0.0 - 1.0) of the generated audio stream. Default value is 1.0.
Specify the duration of the generated audio stream. Not specifying this option results in noise with an infinite length.
Specify the color of noise. Available noise colors are white, pink, brown, blue, violet and velvet. Default color is white.
Specify a value used to seed the PRNG.
Set the number of samples per each output frame, default is 1024.
Set the density (0.0 - 1.0) for the velvet noise generator, default is 0.05.
anoisesrc=d=60:c=pink:r=44100:a=0.5
Generate odd-tap Hilbert transform FIR coefficients.
The resulting stream can be used with afir filter for phase-shifting the signal by 90 degrees.
This is used in many matrix coding schemes and for analytic signal generation. The process is often written as a multiplication by i (or j), the imaginary unit.
The filter accepts the following options:
Set sample rate, default is 44100.
Set length of FIR filter, default is 22051.
Set number of samples per each frame.
Set window function to be used when generating FIR coefficients.
Generate a sinc kaiser-windowed low-pass, high-pass, band-pass, or band-reject FIR coefficients.
The resulting stream can be used with afir filter for filtering the audio signal.
The filter accepts the following options:
Set sample rate, default is 44100.
Set number of samples per each frame. Default is 1024.
Set high-pass frequency. Default is 0.
Set low-pass frequency. Default is 0. If high-pass frequency is lower than low-pass frequency and low-pass frequency is higher than 0 then filter will create band-pass filter coefficients, otherwise band-reject filter coefficients.
Set filter phase response. Default is 50. Allowed range is from 0 to 100.
Set Kaiser window beta.
Set stop-band attenuation. Default is 120dB, allowed range is from 40 to 180 dB.
Enable rounding, by default is disabled.
Set number of taps for high-pass filter.
Set number of taps for low-pass filter.
Generate an audio signal made of a sine wave with amplitude 1/8.
The audio signal is bit-exact.
The filter accepts the following options:
Set the carrier frequency. Default is 440 Hz.
Enable a periodic beep every second with frequency beep_factor times the carrier frequency. Default is 0, meaning the beep is disabled.
Specify the sample rate, default is 44100.
Specify the duration of the generated audio stream.
Set the number of samples per output frame.
The expression can contain the following constants:
The (sequential) number of the output audio frame, starting from 0.
The PTS (Presentation TimeStamp) of the output audio frame, expressed in TB units.
The PTS of the output audio frame, expressed in seconds.
The timebase of the output audio frames.
Default is 1024
.
sine
sine=220:4:d=5 sine=f=220:b=4:d=5 sine=frequency=220:beep_factor=4:duration=5
1602,1601,1602,1601,1602
NTSC
pattern:
sine=1000:samples_per_frame='st(0,mod(n,5)); 1602-not(not(eq(ld(0),1)+eq(ld(0),3)))'
Below is a description of the currently available audio sinks.
Buffer audio frames, and make them available to the end of filter chain.
This sink is mainly intended for programmatic use, in particular through the interface defined in libavfilter/buffersink.h or the options system.
It accepts a pointer to an AVABufferSinkContext structure, which
defines the incoming buffers’ formats, to be passed as the opaque
parameter to avfilter_init_filter
for initialization.
Null audio sink; do absolutely nothing with the input audio. It is mainly useful as a template and for use in analysis / debugging tools.
When you configure your FFmpeg build, you can disable any of the
existing filters using --disable-filters
.
The configure output will show the video filters included in your
build.
Below is a description of the currently available video filters.
Mark a region of interest in a video frame.
The frame data is passed through unchanged, but metadata is attached to the frame indicating regions of interest which can affect the behaviour of later encoding. Multiple regions can be marked by applying the filter multiple times.
Region distance in pixels from the left edge of the frame.
Region distance in pixels from the top edge of the frame.
Region width in pixels.
Region height in pixels.
The parameters x, y, w and h are expressions, and may contain the following variables:
Width of the input frame.
Height of the input frame.
Quantisation offset to apply within the region.
This must be a real value in the range -1 to +1. A value of zero indicates no quality change. A negative value asks for better quality (less quantisation), while a positive value asks for worse quality (greater quantisation).
The range is calibrated so that the extreme values indicate the largest possible offset - if the rest of the frame is encoded with the worst possible quality, an offset of -1 indicates that this region should be encoded with the best possible quality anyway. Intermediate values are then interpolated in some codec-dependent way.
For example, in 10-bit H.264 the quantisation parameter varies between -12 and 51. A typical qoffset value of -1/10 therefore indicates that this region should be encoded with a QP around one-tenth of the full range better than the rest of the frame. So, if most of the frame were to be encoded with a QP of around 30, this region would get a QP of around 24 (an offset of approximately -1/10 * (51 - -12) = -6.3). An extreme value of -1 would indicate that this region should be encoded with the best possible quality regardless of the treatment of the rest of the frame - that is, should be encoded at a QP of -12.
If set to true, remove any existing regions of interest marked on the frame before adding the new one.
addroi=iw/4:ih/4:iw/2:ih/2:-1/10
addroi=0:0:100:ih:+1/5
Extract the alpha component from the input as a grayscale video. This is especially useful with the alphamerge filter.
Add or replace the alpha component of the primary input with the grayscale value of a second input. This is intended for use with alphaextract to allow the transmission or storage of frame sequences that have alpha in a format that doesn’t support an alpha channel.
For example, to reconstruct full frames from a normal YUV-encoded video and a separate video created with alphaextract, you might use:
movie=in_alpha.mkv [alpha]; [in][alpha] alphamerge [out]
Amplify differences between current pixel and pixels of adjacent frames in same pixel location.
This filter accepts the following options:
Set frame radius. Default is 2. Allowed range is from 1 to 63. For example radius of 3 will instruct filter to calculate average of 7 frames.
Set factor to amplify difference. Default is 2. Allowed range is from 0 to 65535.
Set threshold for difference amplification. Any difference greater or equal to this value will not alter source pixel. Default is 10. Allowed range is from 0 to 65535.
Set tolerance for difference amplification. Any difference lower to this value will not alter source pixel. Default is 0. Allowed range is from 0 to 65535.
Set lower limit for changing source pixel. Default is 65535. Allowed range is from 0 to 65535. This option controls maximum possible value that will decrease source pixel value.
Set high limit for changing source pixel. Default is 65535. Allowed range is from 0 to 65535. This option controls maximum possible value that will increase source pixel value.
Set which planes to filter. Default is all. Allowed range is from 0 to 15.
This filter supports the following commands that corresponds to option of same name:
Same as the subtitles filter, except that it doesn’t require libavcodec and libavformat to work. On the other hand, it is limited to ASS (Advanced Substation Alpha) subtitles files.
This filter accepts the following option in addition to the common options from the subtitles filter:
Set the shaping engine
Available values are:
The default libass shaping engine, which is the best available.
Fast, font-agnostic shaper that can do only substitutions
Slower shaper using OpenType for substitutions and positioning
The default is auto
.
Apply an Adaptive Temporal Averaging Denoiser to the video input.
The filter accepts the following options:
Set threshold A for 1st plane. Default is 0.02. Valid range is 0 to 0.3.
Set threshold B for 1st plane. Default is 0.04. Valid range is 0 to 5.
Set threshold A for 2nd plane. Default is 0.02. Valid range is 0 to 0.3.
Set threshold B for 2nd plane. Default is 0.04. Valid range is 0 to 5.
Set threshold A for 3rd plane. Default is 0.02. Valid range is 0 to 0.3.
Set threshold B for 3rd plane. Default is 0.04. Valid range is 0 to 5.
Threshold A is designed to react on abrupt changes in the input signal and threshold B is designed to react on continuous changes in the input signal.
Set number of frames filter will use for averaging. Default is 9. Must be odd number in range [5, 129].
Set what planes of frame filter will use for averaging. Default is all.
Set what variant of algorithm filter will use for averaging. Default is p
parallel.
Alternatively can be set to s
serial.
Parallel can be faster then serial, while other way around is never true. Parallel will abort early on first change being greater then thresholds, while serial will continue processing other side of frames if they are equal or below thresholds.
Set sigma for 1st plane, 2nd plane or 3rd plane. Default is 32767. Valid range is from 0 to 32767. This options controls weight for each pixel in radius defined by size. Default value means every pixel have same weight. Setting this option to 0 effectively disables filtering.
This filter supports same commands as options except option s
.
The command accepts the same syntax of the corresponding option.
Apply average blur filter.
The filter accepts the following options:
Set horizontal radius size.
Set which planes to filter. By default all planes are filtered.
Set vertical radius size, if zero it will be same as sizeX
.
Default is 0
.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Turns a static background into transparency.
The filter accepts the following option:
Threshold for scene change detection.
Similarity percentage with the background.
Set the blend amount for pixels that are not similar.
This filter supports the all above options as commands.
Compute the bounding box for the non-black pixels in the input frame luma plane.
This filter computes the bounding box containing all the pixels with a luma value greater than the minimum allowed value. The parameters describing the bounding box are printed on the filter log.
The filter accepts the following option:
Set the minimal luma value. Default is 16
.
This filter supports the all above options as commands.
Apply bilateral filter, spatial smoothing while preserving edges.
The filter accepts the following options:
Set sigma of gaussian function to calculate spatial weight. Allowed range is 0 to 512. Default is 0.1.
Set sigma of gaussian function to calculate range weight. Allowed range is 0 to 1. Default is 0.1.
Set planes to filter. Default is first only.
This filter supports the all above options as commands.
CUDA accelerated bilateral filter, an edge preserving filter. This filter is mathematically accurate thanks to the use of GPU acceleration. For best output quality, use one to one chroma subsampling, i.e. yuv444p format.
The filter accepts the following options:
Set sigma of gaussian function to calculate spatial weight, also called sigma space. Allowed range is 0.1 to 512. Default is 0.1.
Set sigma of gaussian function to calculate color range weight, also called sigma color. Allowed range is 0.1 to 512. Default is 0.1.
Set window size of the bilateral function to determine the number of neighbours to loop on. If the number entered is even, one will be added automatically. Allowed range is 1 to 255. Default is 1.
./ffmpeg -v verbose \ -hwaccel cuda -hwaccel_output_format cuda -i input.mp4 \ -init_hw_device cuda \ -filter_complex \ " \ [0:v]scale_cuda=format=yuv444p[scaled_video]; [scaled_video]bilateral_cuda=window_size=9:sigmaS=3.0:sigmaR=50.0" \ -an -sn -c:v h264_nvenc -cq 20 out.mp4
Show and measure bit plane noise.
The filter accepts the following options:
Set which plane to analyze. Default is 1
.
Filter out noisy pixels from bitplane
set above.
Default is disabled.
Detect video intervals that are (almost) completely black. Can be useful to detect chapter transitions, commercials, or invalid recordings.
The filter outputs its detection analysis to both the log as well as
frame metadata. If a black segment of at least the specified minimum
duration is found, a line with the start and end timestamps as well
as duration is printed to the log with level info
. In addition,
a log line with level debug
is printed per frame showing the
black amount detected for that frame.
The filter also attaches metadata to the first frame of a black
segment with key lavfi.black_start
and to the first frame
after the black segment ends with key lavfi.black_end
. The
value is the frame’s timestamp. This metadata is added regardless
of the minimum duration specified.
The filter accepts the following options:
Set the minimum detected black duration expressed in seconds. It must be a non-negative floating point number.
Default value is 2.0.
Set the threshold for considering a picture "black". Express the minimum value for the ratio:
nb_black_pixels / nb_pixels
for which a picture is considered black. Default value is 0.98.
Set the threshold for considering a pixel "black".
The threshold expresses the maximum pixel luma value for which a pixel is considered "black". The provided value is scaled according to the following equation:
absolute_threshold = luma_minimum_value + pixel_black_th * luma_range_size
luma_range_size and luma_minimum_value depend on the input video format, the range is [0-255] for YUV full-range formats and [16-235] for YUV non full-range formats.
Default value is 0.10.
The following example sets the maximum pixel threshold to the minimum value, and detects only black intervals of 2 or more seconds:
blackdetect=d=2:pix_th=0.00
Detect frames that are (almost) completely black. Can be useful to detect chapter transitions or commercials. Output lines consist of the frame number of the detected frame, the percentage of blackness, the position in the file if known or -1 and the timestamp in seconds.
In order to display the output lines, you need to set the loglevel at least to the AV_LOG_INFO value.
This filter exports frame metadata lavfi.blackframe.pblack
.
The value represents the percentage of pixels in the picture that
are below the threshold value.
It accepts the following parameters:
The percentage of the pixels that have to be below the threshold; it defaults to
98
.
The threshold below which a pixel value is considered black; it defaults to
32
.
Blend two video frames into each other.
The blend
filter takes two input streams and outputs one
stream, the first input is the "top" layer and second input is
"bottom" layer. By default, the output terminates when the longest input terminates.
The tblend
(time blend) filter takes two consecutive frames
from one single stream, and outputs the result obtained by blending
the new frame on top of the old frame.
A description of the accepted options follows.
Set blend mode for specific pixel component or all pixel components in case
of all_mode. Default value is normal
.
Available values for component modes are:
Set blend opacity for specific pixel component or all pixel components in case of all_opacity. Only used in combination with pixel component blend modes.
Set blend expression for specific pixel component or all pixel components in case of all_expr. Note that related mode options will be ignored if those are set.
The expressions can use the following variables:
The sequential number of the filtered frame, starting from 0
.
the coordinates of the current sample
the width and height of currently filtered plane
Width and height scale for the plane being filtered. It is the
ratio between the dimensions of the current plane to the luma plane,
e.g. for a yuv420p
frame, the values are 1,1
for
the luma plane and 0.5,0.5
for the chroma planes.
Time of the current frame, expressed in seconds.
Value of pixel component at current location for first video frame (top layer).
Value of pixel component at current location for second video frame (bottom layer).
The blend
filter also supports the framesync options.
blend=all_expr='A*(if(gte(T,10),1,T/10))+B*(1-(if(gte(T,10),1,T/10)))'
blend=all_expr='A*(X/W)+B*(1-X/W)'
blend=all_expr='if(eq(mod(X,2),mod(Y,2)),A,B)'
blend=all_expr='if(gte(N*SW+X,W),A,B)'
blend=all_expr='if(gte(Y-N*SH,0),A,B)'
blend=all_expr='if(gte(T*SH*40+Y,H)*gte((T*40*SW+X)*W/H,W),A,B)'
blend=all_expr='if(gt(X,Y*(W/H)),A,B)'
tblend=all_mode=grainextract
Determines blockiness of frames without altering the input frames.
Based on Remco Muijs and Ihor Kirenko: "A no-reference blocking artifact measure for adaptive video processing." 2005 13th European signal processing conference.
The filter accepts the following options:
Set minimum and maximum values for determining pixel grids (periods). Default values are [3,24].
Set planes to filter. Default is first only.
Determines blurriness of frames without altering the input frames.
Based on Marziliano, Pina, et al. "A no-reference perceptual blur metric." Allows for a block-based abbreviation.
The filter accepts the following options:
Set low and high threshold values used by the Canny thresholding algorithm.
The high threshold selects the "strong" edge pixels, which are then connected through 8-connectivity with the "weak" edge pixels selected by the low threshold.
low and high threshold values must be chosen in the range [0,1], and low should be lesser or equal to high.
Default value for low is 20/255
, and default value for high
is 50/255
.
Define the radius to search around an edge pixel for local maxima.
Determine blurriness only for the most significant blocks, given in percentage.
Determine blurriness for blocks of width block_width. If set to any value smaller 1, no blocks are used and the whole image is processed as one no matter of block_height.
Determine blurriness for blocks of height block_height. If set to any value smaller 1, no blocks are used and the whole image is processed as one no matter of block_width.
Set planes to filter. Default is first only.
blurdetect=block_width=32:block_height=32:block_pct=80
Denoise frames using Block-Matching 3D algorithm.
The filter accepts the following options.
Set denoising strength. Default value is 1. Allowed range is from 0 to 999.9. The denoising algorithm is very sensitive to sigma, so adjust it according to the source.
Set local patch size. This sets dimensions in 2D.
Set sliding step for processing blocks. Default value is 4. Allowed range is from 1 to 64. Smaller values allows processing more reference blocks and is slower.
Set maximal number of similar blocks for 3rd dimension. Default value is 1. When set to 1, no block matching is done. Larger values allows more blocks in single group. Allowed range is from 1 to 256.
Set radius for search block matching. Default is 9. Allowed range is from 1 to INT32_MAX.
Set step between two search locations for block matching. Default is 1. Allowed range is from 1 to 64. Smaller is slower.
Set threshold of mean square error for block matching. Valid range is 0 to INT32_MAX.
Set thresholding parameter for hard thresholding in 3D transformed domain. Larger values results in stronger hard-thresholding filtering in frequency domain.
Set filtering estimation mode. Can be basic
or final
.
Default is basic
.
If enabled, filter will use 2nd stream for block matching.
Default is disabled for basic
value of estim option,
and always enabled if value of estim is final
.
Set planes to filter. Default is all available except alpha.
bm3d=sigma=3:block=4:bstep=2:group=1:estim=basic
bm3d=sigma=3:block=4:bstep=2:group=1:estim=basic:planes=1
split[a][b],[a]bm3d=sigma=3:block=4:bstep=2:group=1:estim=basic[a],[b][a]bm3d=sigma=3:block=4:bstep=2:group=16:estim=final:ref=1
split[a][b],[a]nlmeans=s=3:r=7:p=3[a],[b][a]bm3d=sigma=3:block=4:bstep=2:group=16:estim=final:ref=1
Apply a boxblur algorithm to the input video.
It accepts the following parameters:
A description of the accepted options follows.
Set an expression for the box radius in pixels used for blurring the corresponding input plane.
The radius value must be a non-negative number, and must not be
greater than the value of the expression min(w,h)/2
for the
luma and alpha planes, and of min(cw,ch)/2
for the chroma
planes.
Default value for luma_radius is "2". If not specified, chroma_radius and alpha_radius default to the corresponding value set for luma_radius.
The expressions can contain the following constants:
The input width and height in pixels.
The input chroma image width and height in pixels.
The horizontal and vertical chroma subsample values. For example, for the pixel format "yuv422p", hsub is 2 and vsub is 1.
Specify how many times the boxblur filter is applied to the corresponding plane.
Default value for luma_power is 2. If not specified, chroma_power and alpha_power default to the corresponding value set for luma_power.
A value of 0 will disable the effect.
boxblur=luma_radius=2:luma_power=1 boxblur=2:1
boxblur=2:1:cr=0:ar=0
boxblur=luma_radius=min(h\,w)/10:luma_power=1:chroma_radius=min(cw\,ch)/10:chroma_power=1
Deinterlace the input video ("bwdif" stands for "Bob Weaver Deinterlacing Filter").
Motion adaptive deinterlacing based on yadif with the use of w3fdif and cubic interpolation algorithms. It accepts the following parameters:
The interlacing mode to adopt. It accepts one of the following values:
Output one frame for each frame.
Output one frame for each field.
The default value is send_field
.
The picture field parity assumed for the input interlaced video. It accepts one of the following values:
Assume the top field is first.
Assume the bottom field is first.
Enable automatic detection of field parity.
The default value is auto
.
If the interlacing is unknown or the decoder does not export this information,
top field first will be assumed.
Specify which frames to deinterlace. Accepts one of the following values:
Deinterlace all frames.
Only deinterlace frames marked as interlaced.
The default value is all
.
Deinterlace the input video using the bwdif algorithm, but implemented in CUDA so that it can work as part of a GPU accelerated pipeline with nvdec and/or nvenc.
It accepts the following parameters:
The interlacing mode to adopt. It accepts one of the following values:
Output one frame for each frame.
Output one frame for each field.
The default value is send_field
.
The picture field parity assumed for the input interlaced video. It accepts one of the following values:
Assume the top field is first.
Assume the bottom field is first.
Enable automatic detection of field parity.
The default value is auto
.
If the interlacing is unknown or the decoder does not export this information,
top field first will be assumed.
Specify which frames to deinterlace. Accepts one of the following values:
Deinterlace all frames.
Only deinterlace frames marked as interlaced.
The default value is all
.
Repack CEA-708 closed captioning side data
This filter fixes various issues seen with commerical encoders related to upstream malformed CEA-708 payloads, specifically incorrect number of tuples (wrong cc_count for the target FPS), and incorrect ordering of tuples (i.e. the CEA-608 tuples are not at the first entries in the payload).
Apply Contrast Adaptive Sharpen filter to video stream.
The filter accepts the following options:
Set the sharpening strength. Default value is 0.
Set planes to filter. Default value is to filter all planes except alpha plane.
This filter supports same commands as options.
Remove all color information for all colors except for certain one.
The filter accepts the following options:
The color which will not be replaced with neutral chroma.
Similarity percentage with the above color. 0.01 matches only the exact key color, while 1.0 matches everything.
Blend percentage. 0.0 makes pixels either fully gray, or not gray at all. Higher values result in more preserved color.
Signals that the color passed is already in YUV instead of RGB.
Literal colors like "green" or "red" don’t make sense with this enabled anymore. This can be used to pass exact YUV values as hexadecimal numbers.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
YUV colorspace color/chroma keying.
The filter accepts the following options:
The color which will be replaced with transparency.
Similarity percentage with the key color.
0.01 matches only the exact key color, while 1.0 matches everything.
Blend percentage.
0.0 makes pixels either fully transparent, or not transparent at all.
Higher values result in semi-transparent pixels, with a higher transparency the more similar the pixels color is to the key color.
Signals that the color passed is already in YUV instead of RGB.
Literal colors like "green" or "red" don’t make sense with this enabled anymore. This can be used to pass exact YUV values as hexadecimal numbers.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
ffmpeg -i input.png -vf chromakey=green out.png
ffmpeg -f lavfi -i color=c=black:s=1280x720 -i video.mp4 -shortest -filter_complex "[1:v]chromakey=0x70de77:0.1:0.2[ckout];[0:v][ckout]overlay[out]" -map "[out]" output.mkv
CUDA accelerated YUV colorspace color/chroma keying.
This filter works like normal chromakey filter but operates on CUDA frames. for more details and parameters see chromakey.
./ffmpeg \ -hwaccel cuda -hwaccel_output_format cuda -i input_green.mp4 \ -hwaccel cuda -hwaccel_output_format cuda -i base_video.mp4 \ -init_hw_device cuda \ -filter_complex \ " \ [0:v]chromakey_cuda=0x25302D:0.1:0.12:1[overlay_video]; \ [1:v]scale_cuda=format=yuv420p[base]; \ [base][overlay_video]overlay_cuda" \ -an -sn -c:v h264_nvenc -cq 20 output.mp4
./ffmpeg -init_hw_device cuda=cuda -filter_hw_device cuda \ -f lavfi -i color=size=800x600:color=white,format=yuv420p \ -f lavfi -i yuvtestsrc=size=200x200,format=yuv420p \ -filter_complex \ " \ [0]hwupload[under]; \ [1]hwupload,chromakey_cuda=green:0.1:0.12[over]; \ [under][over]overlay_cuda" \ -c:v hevc_nvenc -cq 18 -preset slow output.mp4
Reduce chrominance noise.
The filter accepts the following options:
Set threshold for averaging chrominance values. Sum of absolute difference of Y, U and V pixel components of current pixel and neighbour pixels lower than this threshold will be used in averaging. Luma component is left unchanged and is copied to output. Default value is 30. Allowed range is from 1 to 200.
Set horizontal radius of rectangle used for averaging. Allowed range is from 1 to 100. Default value is 5.
Set vertical radius of rectangle used for averaging. Allowed range is from 1 to 100. Default value is 5.
Set horizontal step when averaging. Default value is 1. Allowed range is from 1 to 50. Mostly useful to speed-up filtering.
Set vertical step when averaging. Default value is 1. Allowed range is from 1 to 50. Mostly useful to speed-up filtering.
Set Y threshold for averaging chrominance values. Set finer control for max allowed difference between Y components of current pixel and neigbour pixels. Default value is 200. Allowed range is from 1 to 200.
Set U threshold for averaging chrominance values. Set finer control for max allowed difference between U components of current pixel and neigbour pixels. Default value is 200. Allowed range is from 1 to 200.
Set V threshold for averaging chrominance values. Set finer control for max allowed difference between V components of current pixel and neigbour pixels. Default value is 200. Allowed range is from 1 to 200.
Set distance type used in calculations.
Absolute difference.
Difference squared.
Default distance type is manhattan.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
Shift chroma pixels horizontally and/or vertically.
The filter accepts the following options:
Set amount to shift chroma-blue horizontally.
Set amount to shift chroma-blue vertically.
Set amount to shift chroma-red horizontally.
Set amount to shift chroma-red vertically.
Set edge mode, can be smear, default, or warp.
This filter supports the all above options as commands.
Display CIE color diagram with pixels overlaid onto it.
The filter accepts the following options:
Set color system.
Set CIE system.
Set what gamuts to draw.
See system
option for available values.
Set ciescope size, by default set to 512.
Set intensity used to map input pixel values to CIE diagram.
Set contrast used to draw tongue colors that are out of active color system gamut.
Correct gamma displayed on scope, by default enabled.
Show white point on CIE diagram, by default disabled.
Set input gamma. Used only with XYZ input color space.
Fill with CIE colors. By default is enabled.
Visualize information exported by some codecs.
Some codecs can export information through frames using side-data or other means. For example, some MPEG based codecs export motion vectors through the export_mvs flag in the codec flags2 option.
The filter accepts the following option:
Display block partition structure using the luma plane.
Set motion vectors to visualize.
Available flags for mv are:
forward predicted MVs of P-frames
forward predicted MVs of B-frames
backward predicted MVs of B-frames
Display quantization parameters using the chroma planes.
Set motion vectors type to visualize. Includes MVs from all frames unless specified by frame_type option.
Available flags for mv_type are:
forward predicted MVs
backward predicted MVs
Set frame type to visualize motion vectors of.
Available flags for frame_type are:
intra-coded frames (I-frames)
predicted frames (P-frames)
bi-directionally predicted frames (B-frames)
ffplay
:
ffplay -flags2 +export_mvs input.mp4 -vf codecview=mv_type=fp
ffplay
:
ffplay -flags2 +export_mvs input.mp4 -vf codecview=mv=pf+bf+bb
Modify intensity of primary colors (red, green and blue) of input frames.
The filter allows an input frame to be adjusted in the shadows, midtones or highlights regions for the red-cyan, green-magenta or blue-yellow balance.
A positive adjustment value shifts the balance towards the primary color, a negative value towards the complementary color.
The filter accepts the following options:
Adjust red, green and blue shadows (darkest pixels).
Adjust red, green and blue midtones (medium pixels).
Adjust red, green and blue highlights (brightest pixels).
Allowed ranges for options are [-1.0, 1.0]
. Defaults are 0
.
Preserve lightness when changing color balance. Default is disabled.
colorbalance=rs=.3
This filter supports the all above options as commands.
Adjust color contrast between RGB components.
The filter accepts the following options:
Set the red-cyan contrast. Defaults is 0.0. Allowed range is from -1.0 to 1.0.
Set the green-magenta contrast. Defaults is 0.0. Allowed range is from -1.0 to 1.0.
Set the blue-yellow contrast. Defaults is 0.0. Allowed range is from -1.0 to 1.0.
Set the weight of each rc
, gm
, by
option value. Default value is 0.0.
Allowed range is from 0.0 to 1.0. If all weights are 0.0 filtering is disabled.
Set the amount of preserving lightness. Default value is 0.0. Allowed range is from 0.0 to 1.0.
This filter supports the all above options as commands.
Adjust color white balance selectively for blacks and whites. This filter operates in YUV colorspace.
The filter accepts the following options:
Set the red shadow spot. Allowed range is from -1.0 to 1.0. Default value is 0.
Set the blue shadow spot. Allowed range is from -1.0 to 1.0. Default value is 0.
Set the red highlight spot. Allowed range is from -1.0 to 1.0. Default value is 0.
Set the blue highlight spot. Allowed range is from -1.0 to 1.0. Default value is 0.
Set the amount of saturation. Allowed range is from -3.0 to 3.0. Default value is 1.
If set to anything other than manual
it will analyze every frame and use derived
parameters for filtering output frame.
Possible values are:
Default value is manual
.
This filter supports the all above options as commands.
Adjust video input frames by re-mixing color channels.
This filter modifies a color channel by adding the values associated to the other channels of the same pixels. For example if the value to modify is red, the output value will be:
red=red*rr + blue*rb + green*rg + alpha*ra
The filter accepts the following options:
Adjust contribution of input red, green, blue and alpha channels for output red channel.
Default is 1
for rr, and 0
for rg, rb and ra.
Adjust contribution of input red, green, blue and alpha channels for output green channel.
Default is 1
for gg, and 0
for gr, gb and ga.
Adjust contribution of input red, green, blue and alpha channels for output blue channel.
Default is 1
for bb, and 0
for br, bg and ba.
Adjust contribution of input red, green, blue and alpha channels for output alpha channel.
Default is 1
for aa, and 0
for ar, ag and ab.
Allowed ranges for options are [-2.0, 2.0]
.
Set preserve color mode. The accepted values are:
Disable color preserving, this is default.
Preserve luminance.
Preserve max value of RGB triplet.
Preserve average value of RGB triplet.
Preserve sum value of RGB triplet.
Preserve normalized value of RGB triplet.
Preserve power value of RGB triplet.
Set the preserve color amount when changing colors. Allowed range is from [0.0, 1.0]
.
Default is 0.0
, thus disabled.
colorchannelmixer=.3:.4:.3:0:.3:.4:.3:0:.3:.4:.3
colorchannelmixer=.393:.769:.189:0:.349:.686:.168:0:.272:.534:.131
This filter supports the all above options as commands.
Overlay a solid color on the video stream.
The filter accepts the following options:
Set the color hue. Allowed range is from 0 to 360. Default value is 0.
Set the color saturation. Allowed range is from 0 to 1. Default value is 0.5.
Set the color lightness. Allowed range is from 0 to 1. Default value is 0.5.
Set the mix of source lightness. By default is set to 1.0. Allowed range is from 0.0 to 1.0.
This filter supports the all above options as commands.
RGB colorspace color keying. This filter operates on 8-bit RGB format frames by setting the alpha component of each pixel which falls within the similarity radius of the key color to 0. The alpha value for pixels outside the similarity radius depends on the value of the blend option.
The filter accepts the following options:
Set the color for which alpha will be set to 0 (full transparency).
See (ffmpeg-utils)"Color" section in the ffmpeg-utils manual.
Default is black
.
Set the radius from the key color within which other colors also have full transparency.
The computed distance is related to the unit fractional distance in 3D space between the RGB values
of the key color and the pixel’s color. Range is 0.01 to 1.0. 0.01 matches within a very small radius
around the exact key color, while 1.0 matches everything.
Default is 0.01
.
Set how the alpha value for pixels that fall outside the similarity radius is computed.
0.0 makes pixels either fully transparent or fully opaque.
Higher values result in semi-transparent pixels, with greater transparency
the more similar the pixel color is to the key color.
Range is 0.0 to 1.0. Default is 0.0
.
ffmpeg -i input.png -vf colorkey=green out.png
ffmpeg -i background.png -i video.mp4 -filter_complex "[1:v]colorkey=0x3BBD1E:0.3:0.2[ckout];[0:v][ckout]overlay[out]" -map "[out]" output.flv
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Remove all color information for all RGB colors except for certain one.
The filter accepts the following options:
The color which will not be replaced with neutral gray.
Similarity percentage with the above color. 0.01 matches only the exact key color, while 1.0 matches everything.
Blend percentage. 0.0 makes pixels fully gray. Higher values result in more preserved color.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Adjust video input frames using levels.
The filter accepts the following options:
Adjust red, green, blue and alpha input black point.
Allowed ranges for options are [-1.0, 1.0]
. Defaults are 0
.
Adjust red, green, blue and alpha input white point.
Allowed ranges for options are [-1.0, 1.0]
. Defaults are 1
.
Input levels are used to lighten highlights (bright tones), darken shadows (dark tones), change the balance of bright and dark tones.
Adjust red, green, blue and alpha output black point.
Allowed ranges for options are [0, 1.0]
. Defaults are 0
.
Adjust red, green, blue and alpha output white point.
Allowed ranges for options are [0, 1.0]
. Defaults are 1
.
Output levels allows manual selection of a constrained output level range.
Set preserve color mode. The accepted values are:
Disable color preserving, this is default.
Preserve luminance.
Preserve max value of RGB triplet.
Preserve average value of RGB triplet.
Preserve sum value of RGB triplet.
Preserve normalized value of RGB triplet.
Preserve power value of RGB triplet.
colorlevels=rimin=0.058:gimin=0.058:bimin=0.058
colorlevels=rimin=0.039:gimin=0.039:bimin=0.039:rimax=0.96:gimax=0.96:bimax=0.96
colorlevels=rimax=0.902:gimax=0.902:bimax=0.902
colorlevels=romin=0.5:gomin=0.5:bomin=0.5
This filter supports the all above options as commands.
Apply custom color maps to video stream.
This filter needs three input video streams. First stream is video stream that is going to be filtered out. Second and third video stream specify color patches for source color to target color mapping.
The filter accepts the following options:
Set the source and target video stream patch size in pixels.
Set the max number of used patches from source and target video stream.
Default value is number of patches available in additional video streams.
Max allowed number of patches is 64
.
Set the adjustments used for target colors. Can be relative
or absolute
.
Defaults is absolute
.
Set the kernel used to measure color differences between mapped colors.
The accepted values are:
Default is euclidean
.
Convert color matrix.
The filter accepts the following options:
Specify the source and destination color matrix. Both values must be specified.
The accepted values are:
BT.709
FCC
BT.601
BT.470
BT.470BG
SMPTE-170M
SMPTE-240M
BT.2020
For example to convert from BT.601 to SMPTE-240M, use the command:
colormatrix=bt601:smpte240m
Convert colorspace, transfer characteristics or color primaries. Input video needs to have an even size.
The filter accepts the following options:
Specify all color properties at once.
The accepted values are:
BT.470M
BT.470BG
BT.601-6 525
BT.601-6 625
BT.709
SMPTE-170M
SMPTE-240M
BT.2020
Specify output colorspace.
The accepted values are:
BT.709
FCC
BT.470BG or BT.601-6 625
SMPTE-170M or BT.601-6 525
SMPTE-240M
YCgCo
BT.2020 with non-constant luminance
Specify output transfer characteristics.
The accepted values are:
BT.709
BT.470M
BT.470BG
Constant gamma of 2.2
Constant gamma of 2.8
SMPTE-170M, BT.601-6 625 or BT.601-6 525
SMPTE-240M
SRGB
iec61966-2-1
iec61966-2-4
xvycc
BT.2020 for 10-bits content
BT.2020 for 12-bits content
Specify output color primaries.
The accepted values are:
BT.709
BT.470M
BT.470BG or BT.601-6 625
SMPTE-170M or BT.601-6 525
SMPTE-240M
film
SMPTE-431
SMPTE-432
BT.2020
JEDEC P22 phosphors
Specify output color range.
The accepted values are:
TV (restricted) range
MPEG (restricted) range
PC (full) range
JPEG (full) range
Specify output color format.
The accepted values are:
YUV 4:2:0 planar 8-bits
YUV 4:2:0 planar 10-bits
YUV 4:2:0 planar 12-bits
YUV 4:2:2 planar 8-bits
YUV 4:2:2 planar 10-bits
YUV 4:2:2 planar 12-bits
YUV 4:4:4 planar 8-bits
YUV 4:4:4 planar 10-bits
YUV 4:4:4 planar 12-bits
Do a fast conversion, which skips gamma/primary correction. This will take significantly less CPU, but will be mathematically incorrect. To get output compatible with that produced by the colormatrix filter, use fast=1.
Specify dithering mode.
The accepted values are:
No dithering
Floyd-Steinberg dithering
Whitepoint adaptation mode.
The accepted values are:
Bradford whitepoint adaptation
von Kries whitepoint adaptation
identity whitepoint adaptation (i.e. no whitepoint adaptation)
Override all input properties at once. Same accepted values as all.
Override input colorspace. Same accepted values as space.
Override input color primaries. Same accepted values as primaries.
Override input transfer characteristics. Same accepted values as trc.
Override input color range. Same accepted values as range.
The filter converts the transfer characteristics, color space and color primaries to the specified user values. The output value, if not specified, is set to a default value based on the "all" property. If that property is also not specified, the filter will log an error. The output color range and format default to the same value as the input color range and format. The input transfer characteristics, color space, color primaries and color range should be set on the input data. If any of these are missing, the filter will log an error and no conversion will take place.
For example to convert the input to SMPTE-240M, use the command:
colorspace=smpte240m
CUDA accelerated implementation of the colorspace filter.
It is by no means feature complete compared to the software colorspace filter, and at the current time only supports color range conversion between jpeg/full and mpeg/limited range.
The filter accepts the following options:
Specify output color range.
The accepted values are:
TV (restricted) range
MPEG (restricted) range
PC (full) range
JPEG (full) range
Adjust color temperature in video to simulate variations in ambient color temperature.
The filter accepts the following options:
Set the temperature in Kelvin. Allowed range is from 1000 to 40000. Default value is 6500 K.
Set mixing with filtered output. Allowed range is from 0 to 1. Default value is 1.
Set the amount of preserving lightness. Allowed range is from 0 to 1. Default value is 0.
This filter supports same commands as options.
Apply convolution of 3x3, 5x5, 7x7 or horizontal/vertical up to 49 elements.
The filter accepts the following options:
Set matrix for each plane. Matrix is sequence of 9, 25 or 49 signed integers in square mode, and from 1 to 49 odd number of signed integers in row mode.
Set multiplier for calculated value for each plane. If unset or 0, it will be sum of all matrix elements.
Set bias for each plane. This value is added to the result of the multiplication. Useful for making the overall image brighter or darker. Default is 0.0.
Set matrix mode for each plane. Can be square, row or column. Default is square.
This filter supports the all above options as commands.
convolution="0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0"
convolution="1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1/9:1/9:1/9:1/9"
convolution="0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:5:1:1:1:0:128:128:128"
convolution="0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:5:5:5:1:0:128:128:128"
convolution="1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:5:5:5:1:0:128:128:0"
convolution="-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2"
Apply 2D convolution of video stream in frequency domain using second stream as impulse.
The filter accepts the following options:
Set which planes to process.
Set which impulse video frames will be processed, can be first or all. Default is all.
The convolve
filter also supports the framesync options.
Copy the input video source unchanged to the output. This is mainly useful for testing purposes.
Video filtering on GPU using Apple’s CoreImage API on OSX.
Hardware acceleration is based on an OpenGL context. Usually, this means it is processed by video hardware. However, software-based OpenGL implementations exist which means there is no guarantee for hardware processing. It depends on the respective OSX.
There are many filters and image generators provided by Apple that come with a large variety of options. The filter has to be referenced by its name along with its options.
The coreimage filter accepts the following options:
List all available filters and generators along with all their respective options as well as possible minimum and maximum values along with the default values.
list_filters=true
Specify all filters by their respective name and options.
Use list_filters to determine all valid filter names and options.
Numerical options are specified by a float value and are automatically clamped
to their respective value range. Vector and color options have to be specified
by a list of space separated float values. Character escaping has to be done.
A special option name default
is available to use default options for a
filter.
It is required to specify either default
or at least one of the filter options.
All omitted options are used with their default values.
The syntax of the filter string is as follows:
filter=<NAME>@<OPTION>=<VALUE>[@<OPTION>=<VALUE>][@...][#<NAME>@<OPTION>=<VALUE>[@<OPTION>=<VALUE>][@...]][#...]
Specify a rectangle where the output of the filter chain is copied into the input image. It is given by a list of space separated float values:
output_rect=x\ y\ width\ height
If not given, the output rectangle equals the dimensions of the input image. The output rectangle is automatically cropped at the borders of the input image. Negative values are valid for each component.
output_rect=25\ 25\ 100\ 100
Several filters can be chained for successive processing without GPU-HOST transfers allowing for fast processing of complex filter chains. Currently, only filters with zero (generators) or exactly one (filters) input image and one output image are supported. Also, transition filters are not yet usable as intended.
Some filters generate output images with additional padding depending on the respective filter kernel. The padding is automatically removed to ensure the filter output has the same size as the input image.
For image generators, the size of the output image is determined by the previous output image of the filter chain or the input image of the whole filterchain, respectively. The generators do not use the pixel information of this image to generate their output. However, the generated output is blended onto this image, resulting in partial or complete coverage of the output image.
The coreimagesrc video source can be used for generating input images which are directly fed into the filter chain. By using it, providing input images by another video source or an input video is not required.
coreimage=list_filters=true
coreimage=filter=CIBoxBlur@default
coreimage=filter=CIBoxBlur@default#CIVignetteEffect@inputCenter=100\ 100@inputRadius=50
ffmpeg -f lavfi -i nullsrc=s=100x100,coreimage=filter=CIQRCodeGenerator@inputMessage=https\\\\\://FFmpeg.org/@inputCorrectionLevel=H -frames:v 1 QRCode.png
Obtain the correlation between two input videos.
This filter takes two input videos.
Both input videos must have the same resolution and pixel format for this filter to work correctly. Also it assumes that both inputs have the same number of frames, which are compared one by one.
The obtained per component, average, min and max correlation is printed through the logging system.
The filter stores the calculated correlation of each frame in frame metadata.
This filter also supports the framesync options.
In the below example the input file main.mpg being processed is compared with the reference file ref.mpg.
ffmpeg -i main.mpg -i ref.mpg -lavfi corr -f null -
Cover a rectangular object
It accepts the following options:
Filepath of the optional cover image, needs to be in yuv420.
Set covering mode.
It accepts the following values:
cover it by the supplied image
cover it by interpolating the surrounding pixels
Default value is blur.
ffmpeg
:
ffmpeg -i file.ts -vf find_rect=newref.pgm,cover_rect=cover.jpg:mode=cover new.mkv
Crop the input video to given dimensions.
It accepts the following parameters:
The width of the output video. It defaults to iw
.
This expression is evaluated only once during the filter
configuration, or when the ‘w’ or ‘out_w’ command is sent.
The height of the output video. It defaults to ih
.
This expression is evaluated only once during the filter
configuration, or when the ‘h’ or ‘out_h’ command is sent.
The horizontal position, in the input video, of the left edge of the output
video. It defaults to (in_w-out_w)/2
.
This expression is evaluated per-frame.
The vertical position, in the input video, of the top edge of the output video.
It defaults to (in_h-out_h)/2
.
This expression is evaluated per-frame.
If set to 1 will force the output display aspect ratio to be the same of the input, by changing the output sample aspect ratio. It defaults to 0.
Enable exact cropping. If enabled, subsampled videos will be cropped at exact width/height/x/y as specified and will not be rounded to nearest smaller value. It defaults to 0.
The out_w, out_h, x, y parameters are expressions containing the following constants:
The computed values for x and y. They are evaluated for each new frame.
The input width and height.
These are the same as in_w and in_h.
The output (cropped) width and height.
These are the same as out_w and out_h.
same as iw / ih
input sample aspect ratio
input display aspect ratio, it is the same as (iw / ih) * sar
horizontal and vertical chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
The number of the input frame, starting from 0.
the position in the file of the input frame, NAN if unknown; deprecated, do not use
The timestamp expressed in seconds. It’s NAN if the input timestamp is unknown.
The expression for out_w may depend on the value of out_h, and the expression for out_h may depend on out_w, but they cannot depend on x and y, as x and y are evaluated after out_w and out_h.
The x and y parameters specify the expressions for the position of the top-left corner of the output (non-cropped) area. They are evaluated for each frame. If the evaluated value is not valid, it is approximated to the nearest valid value.
The expression for x may depend on y, and the expression for y may depend on x.
crop=100:100:12:34
Using named options, the example above becomes:
crop=w=100:h=100:x=12:y=34
crop=100:100
crop=2/3*in_w:2/3*in_h
crop=out_w=in_h crop=in_h
crop=in_w-100:in_h-100:100:100
crop=in_w-2*10:in_h-2*20
crop=in_w/2:in_h/2:in_w/2:in_h/2
crop=in_w:1/PHI*in_w
crop=in_w/2:in_h/2:(in_w-out_w)/2+((in_w-out_w)/2)*sin(n/10):(in_h-out_h)/2 +((in_h-out_h)/2)*sin(n/7)
crop=in_w/2:in_h/2:(in_w-out_w)/2+((in_w-out_w)/2)*sin(t*10):(in_h-out_h)/2 +((in_h-out_h)/2)*sin(t*13)
crop=in_w/2:in_h/2:y:10+10*sin(n/10)
This filter supports the following commands:
Set width/height of the output video and the horizontal/vertical position in the input video. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Auto-detect the crop size.
It calculates the necessary cropping parameters and prints the recommended parameters via the logging system. The detected dimensions correspond to the non-black or video area of the input video according to mode.
It accepts the following parameters:
Depending on mode crop detection is based on either the mere black value of surrounding pixels or a combination of motion vectors and edge pixels.
Detect black pixels surrounding the playing video. For fine control use option limit.
Detect the playing video by the motion vectors inside the video and scanning for edge pixels typically forming the border of a playing video.
Set higher black value threshold, which can be optionally specified from nothing (0) to everything (255 for 8-bit based formats). An intensity value greater to the set value is considered non-black. It defaults to 24. You can also specify a value between 0.0 and 1.0 which will be scaled depending on the bitdepth of the pixel format.
The value which the width/height should be divisible by. It defaults to 16. The offset is automatically adjusted to center the video. Use 2 to get only even dimensions (needed for 4:2:2 video). 16 is best when encoding to most video codecs.
Set the number of initial frames for which evaluation is skipped. Default is 2. Range is 0 to INT_MAX.
Set the counter that determines after how many frames cropdetect will reset the previously detected largest video area and start over to detect the current optimal crop area. Default value is 0.
This can be useful when channel logos distort the video area. 0 indicates ’never reset’, and returns the largest area encountered during playback.
Set motion in pixel units as threshold for motion detection. It defaults to 8.
Set low and high threshold values used by the Canny thresholding algorithm.
The high threshold selects the "strong" edge pixels, which are then connected through 8-connectivity with the "weak" edge pixels selected by the low threshold.
low and high threshold values must be chosen in the range [0,1], and low should be lesser or equal to high.
Default value for low is 5/255
, and default value for high
is 15/255
.
ffmpeg -i file.mp4 -vf cropdetect,metadata=mode=print -f null -
ffmpeg -i file.mp4 -vf mestimate,cropdetect=mode=mvedges,metadata=mode=print -f null -
ffmpeg -flags2 +export_mvs -i file.mp4 -vf cropdetect=mode=mvedges,metadata=mode=print -f null -
Delay video filtering until a given wallclock timestamp. The filter first passes on preroll amount of frames, then it buffers at most buffer amount of frames and waits for the cue. After reaching the cue it forwards the buffered frames and also any subsequent frames coming in its input.
The filter can be used synchronize the output of multiple ffmpeg processes for realtime output devices like decklink. By putting the delay in the filtering chain and pre-buffering frames the process can pass on data to output almost immediately after the target wallclock timestamp is reached.
Perfect frame accuracy cannot be guaranteed, but the result is good enough for some use cases.
The cue timestamp expressed in a UNIX timestamp in microseconds. Default is 0.
The duration of content to pass on as preroll expressed in seconds. Default is 0.
The maximum duration of content to buffer before waiting for the cue expressed in seconds. Default is 0.
Apply color adjustments using curves.
This filter is similar to the Adobe Photoshop and GIMP curves tools. Each component (red, green and blue) has its values defined by N key points tied from each other using a smooth curve. The x-axis represents the pixel values from the input frame, and the y-axis the new pixel values to be set for the output frame.
By default, a component curve is defined by the two points (0;0) and (1;1). This creates a straight line where each original pixel value is "adjusted" to its own value, which means no change to the image.
The filter allows you to redefine these two points and add some more. A new
curve will be define to pass smoothly through all these new coordinates. The
new defined points needs to be strictly increasing over the x-axis, and their
x and y values must be in the [0;1] interval. The curve is
formed by using a natural or monotonic cubic spline interpolation, depending
on the interp option (default: natural
). The natural
spline produces a smoother curve in general while the monotonic (pchip
)
spline guarantees the transitions between the specified points to be
monotonic. If the computed curves happened to go outside the vector spaces,
the values will be clipped accordingly.
The filter accepts the following options:
Select one of the available color presets. This option can be used in addition to the r, g, b parameters; in this case, the later options takes priority on the preset values. Available presets are:
Default is none
.
Set the master key points. These points will define a second pass mapping. It is sometimes called a "luminance" or "value" mapping. It can be used with r, g, b or all since it acts like a post-processing LUT.
Set the key points for the red component.
Set the key points for the green component.
Set the key points for the blue component.
Set the key points for all components (not including master). Can be used in addition to the other key points component options. In this case, the unset component(s) will fallback on this all setting.
Specify a Photoshop curves file (.acv
) to import the settings from.
Save Gnuplot script of the curves in specified file.
Specify the kind of interpolation. Available algorithms are:
Natural cubic spline using a piece-wise cubic polynomial that is twice continuously differentiable.
Monotonic cubic spline using a piecewise cubic Hermite interpolating polynomial (PCHIP).
To avoid some filtergraph syntax conflicts, each key points list need to be
defined using the following syntax: x0/y0 x1/y1 x2/y2 ...
.
This filter supports same commands as options.
curves=blue='0/0 0.5/0.58 1/1'
curves=r='0/0.11 .42/.51 1/0.95':g='0/0 0.50/0.48 1/1':b='0/0.22 .49/.44 1/0.8'
Here we obtain the following coordinates for each components:
(0;0.11) (0.42;0.51) (1;0.95)
(0;0) (0.50;0.48) (1;1)
(0;0.22) (0.49;0.44) (1;0.80)
curves=preset=vintage
curves=vintage
curves=psfile='MyCurvesPresets/purple.acv':green='0/0 0.45/0.53 1/1'
cross_process
profile using ffmpeg
and gnuplot
:
ffmpeg -f lavfi -i color -vf curves=cross_process:plot=/tmp/curves.plt -frames:v 1 -f null - gnuplot -p /tmp/curves.plt
Video data analysis filter.
This filter shows hexadecimal pixel values of part of video.
The filter accepts the following options:
Set output video size.
Set x offset from where to pick pixels.
Set y offset from where to pick pixels.
Set scope mode, can be one of the following:
Draw hexadecimal pixel values with white color on black background.
Draw hexadecimal pixel values with input video pixel color on black background.
Draw hexadecimal pixel values on color background picked from input video, the text color is picked in such way so its always visible.
Draw rows and columns numbers on left and top of video.
Set background opacity.
Set display number format. Can be hex
, or dec
. Default is hex
.
Set pixel components to display. By default all pixel components are displayed.
This filter supports same commands as options excluding size
option.
Apply Directional blur filter.
The filter accepts the following options:
Set angle of directional blur. Default is 45
.
Set radius of directional blur. Default is 5
.
Set which planes to filter. By default all planes are filtered.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Denoise frames using 2D DCT (frequency domain filtering).
This filter is not designed for real time.
The filter accepts the following options:
Set the noise sigma constant.
This sigma defines a hard threshold of 3 * sigma
; every DCT
coefficient (absolute value) below this threshold with be dropped.
If you need a more advanced filtering, see expr.
Default is 0
.
Set number overlapping pixels for each block. Since the filter can be slow, you may want to reduce this value, at the cost of a less effective filter and the risk of various artefacts.
If the overlapping value doesn’t permit processing the whole input width or height, a warning will be displayed and according borders won’t be denoised.
Default value is blocksize-1, which is the best possible setting.
Set the coefficient factor expression.
For each coefficient of a DCT block, this expression will be evaluated as a multiplier value for the coefficient.
If this is option is set, the sigma option will be ignored.
The absolute value of the coefficient can be accessed through the c variable.
Set the blocksize using the number of bits. 1<<n
defines the
blocksize, which is the width and height of the processed blocks.
The default value is 3 (8x8) and can be raised to 4 for a blocksize of 16x16. Note that changing this setting has huge consequences on the speed processing. Also, a larger block size does not necessarily means a better de-noising.
Apply a denoise with a sigma of 4.5
:
dctdnoiz=4.5
The same operation can be achieved using the expression system:
dctdnoiz=e='gte(c, 4.5*3)'
Violent denoise using a block size of 16x16
:
dctdnoiz=15:n=4
Remove banding artifacts from input video. It works by replacing banded pixels with average value of referenced pixels.
The filter accepts the following options:
Set banding detection threshold for each plane. Default is 0.02. Valid range is 0.00003 to 0.5. If difference between current pixel and reference pixel is less than threshold, it will be considered as banded.
Banding detection range in pixels. Default is 16. If positive, random number in range 0 to set value will be used. If negative, exact absolute value will be used. The range defines square of four pixels around current pixel.
Set direction in radians from which four pixel will be compared. If positive, random direction from 0 to set direction will be picked. If negative, exact of absolute value will be picked. For example direction 0, -PI or -2*PI radians will pick only pixels on same row and -PI/2 will pick only pixels on same column.
If enabled, current pixel is compared with average value of all four surrounding pixels. The default is enabled. If disabled current pixel is compared with all four surrounding pixels. The pixel is considered banded if only all four differences with surrounding pixels are less than threshold.
If enabled, current pixel is changed if and only if all pixel components are banded, e.g. banding detection threshold is triggered for all color components. The default is disabled.
This filter supports the all above options as commands.
Remove blocking artifacts from input video.
The filter accepts the following options:
Set filter type, can be weak or strong. Default is strong. This controls what kind of deblocking is applied.
Set size of block, allowed range is from 4 to 512. Default is 8.
Set blocking detection thresholds. Allowed range is 0 to 1. Defaults are: 0.098 for alpha and 0.05 for the rest. Using higher threshold gives more deblocking strength. Setting alpha controls threshold detection at exact edge of block. Remaining options controls threshold detection near the edge. Each one for below/above or left/right. Setting any of those to 0 disables deblocking.
Set planes to filter. Default is to filter all available planes.
deblock=filter=weak:block=4
deblock=filter=strong:block=4:alpha=0.12:beta=0.07:gamma=0.06:delta=0.05
deblock=filter=strong:block=4:alpha=0.12:beta=0.07:gamma=0.06:delta=0.05:planes=1
deblock=filter=strong:block=4:alpha=0.12:beta=0.07:gamma=0.06:delta=0.05:planes=6
Drop duplicated frames at regular intervals.
The filter accepts the following options:
Set the number of frames from which one will be dropped. Setting this to
N means one frame in every batch of N frames will be dropped.
Default is 5
.
Set the threshold for duplicate detection. If the difference metric for a frame
is less than or equal to this value, then it is declared as duplicate. Default
is 1.1
Set scene change threshold. Default is 15
.
Set the size of the x and y-axis blocks used during metric calculations.
Larger blocks give better noise suppression, but also give worse detection of
small movements. Must be a power of two. Default is 32
.
Mark main input as a pre-processed input and activate clean source input
stream. This allows the input to be pre-processed with various filters to help
the metrics calculation while keeping the frame selection lossless. When set to
1
, the first stream is for the pre-processed input, and the second
stream is the clean source from where the kept frames are chosen. Default is
0
.
Set whether or not chroma is considered in the metric calculations. Default is
1
.
Set whether or not the input only partially contains content to be decimated.
Default is false
.
If enabled video output stream will be in variable frame rate.
Apply 2D deconvolution of video stream in frequency domain using second stream as impulse.
The filter accepts the following options:
Set which planes to process.
Set which impulse video frames will be processed, can be first or all. Default is all.
Set noise when doing divisions. Default is 0.0000001. Useful when width and height are not same and not power of 2 or if stream prior to convolving had noise.
The deconvolve
filter also supports the framesync options.
Reduce cross-luminance (dot-crawl) and cross-color (rainbows) from video.
It accepts the following options:
Set mode of operation. Can be combination of dotcrawl for cross-luminance reduction and/or rainbows for cross-color reduction.
Set spatial luma threshold. Lower values increases reduction of cross-luminance.
Set tolerance for temporal luma. Higher values increases reduction of cross-luminance.
Set tolerance for chroma temporal variation. Higher values increases reduction of cross-color.
Set temporal chroma threshold. Lower values increases reduction of cross-color.
Apply deflate effect to the video.
This filter replaces the pixel by the local(3x3) average by taking into account only values lower than the pixel.
It accepts the following options:
Limit the maximum change for each plane, default is 65535. If 0, plane will remain unchanged.
This filter supports the all above options as commands.
Remove temporal frame luminance variations.
It accepts the following options:
Set moving-average filter size in frames. Default is 5. Allowed range is 2 - 129.
Set averaging mode to smooth temporal luminance variations.
Available values are:
Arithmetic mean
Geometric mean
Harmonic mean
Quadratic mean
Cubic mean
Power mean
Median
Do not actually modify frame. Useful when one only wants metadata.
Remove judder produced by partially interlaced telecined content.
Judder can be introduced, for instance, by pullup filter. If the original
source was partially telecined content then the output of pullup,dejudder
will have a variable frame rate. May change the recorded frame rate of the
container. Aside from that change, this filter will not affect constant frame
rate video.
The option available in this filter is:
Specify the length of the window over which the judder repeats.
Accepts any integer greater than 1. Useful values are:
If the original was telecined from 24 to 30 fps (Film to NTSC).
If the original was telecined from 25 to 30 fps (PAL to NTSC).
If a mixture of the two.
The default is ‘4’.
Suppress a TV station logo by a simple interpolation of the surrounding pixels. Just set a rectangle covering the logo and watch it disappear (and sometimes something even uglier appear - your mileage may vary).
It accepts the following parameters:
Specify the top left corner coordinates of the logo. They must be specified.
Specify the width and height of the logo to clear. They must be specified.
When set to 1, a green rectangle is drawn on the screen to simplify finding the right x, y, w, and h parameters. The default value is 0.
The rectangle is drawn on the outermost pixels which will be (partly) replaced with interpolated values. The values of the next pixels immediately outside this rectangle in each direction will be used to compute the interpolated pixel values inside the rectangle.
Remove the rain in the input image/video by applying the derain methods based on convolutional neural networks. Supported models:
Training as well as model generation scripts are provided in the repository at https://github.com/XueweiMeng/derain_filter.git.
The filter accepts the following options:
Specify which filter to use. This option accepts the following values:
Derain filter. To conduct derain filter, you need to use a derain model.
Dehaze filter. To conduct dehaze filter, you need to use a dehaze model.
Default value is ‘derain’.
Specify which DNN backend to use for model loading and execution. This option accepts the following values:
TensorFlow backend. To enable this backend you
need to install the TensorFlow for C library (see
https://www.tensorflow.org/install/lang_c) and configure FFmpeg with
--enable-libtensorflow
Set path to model file specifying network architecture and its parameters. Note that different backends use different file formats. TensorFlow can load files for only its format.
To get full functionality (such as async execution), please use the dnn_processing filter.
Attempt to fix small changes in horizontal and/or vertical shift. This filter helps remove camera shake from hand-holding a camera, bumping a tripod, moving on a vehicle, etc.
The filter accepts the following options:
Specify a rectangular area where to limit the search for motion vectors. If desired the search for motion vectors can be limited to a rectangular area of the frame defined by its top left corner, width and height. These parameters have the same meaning as the drawbox filter which can be used to visualise the position of the bounding box.
This is useful when simultaneous movement of subjects within the frame might be confused for camera motion by the motion vector search.
If any or all of x, y, w and h are set to -1 then the full frame is used. This allows later options to be set without specifying the bounding box for the motion vector search.
Default - search the whole frame.
Specify the maximum extent of movement in x and y directions in the range 0-64 pixels. Default 16.
Specify how to generate pixels to fill blanks at the edge of the frame. Available values are:
Fill zeroes at blank locations
Original image at blank locations
Extruded edge value at blank locations
Mirrored edge at blank locations
Default value is ‘mirror’.
Specify the blocksize to use for motion search. Range 4-128 pixels, default 8.
Specify the contrast threshold for blocks. Only blocks with more than the specified contrast (difference between darkest and lightest pixels) will be considered. Range 1-255, default 125.
Specify the search strategy. Available values are:
Set exhaustive search
Set less exhaustive search.
Default value is ‘exhaustive’.
If set then a detailed log of the motion search is written to the specified file.
Remove unwanted contamination of foreground colors, caused by reflected color of greenscreen or bluescreen.
This filter accepts the following options:
Set what type of despill to use.
Set how spillmap will be generated.
Set how much to get rid of still remaining spill.
Controls amount of red in spill area.
Controls amount of green in spill area. Should be -1 for greenscreen.
Controls amount of blue in spill area. Should be -1 for bluescreen.
Controls brightness of spill area, preserving colors.
Modify alpha from generated spillmap.
This filter supports the all above options as commands.
Apply an exact inverse of the telecine operation. It requires a predefined pattern specified using the pattern option which must be the same as that passed to the telecine filter.
This filter accepts the following options:
top field first
bottom field first
The default value is top
.
A string of numbers representing the pulldown pattern you wish to apply.
The default value is 23
.
A number representing position of the first frame with respect to the telecine
pattern. This is to be used if the stream is cut. The default value is 0
.
Apply dilation effect to the video.
This filter replaces the pixel by the local(3x3) maximum.
It accepts the following options:
Limit the maximum change for each plane, default is 65535. If 0, plane will remain unchanged.
Flag which specifies the pixel to refer to. Default is 255 i.e. all eight pixels are used.
Flags to local 3x3 coordinates maps like this:
1 2 3 4 5 6 7 8
This filter supports the all above options as commands.
Displace pixels as indicated by second and third input stream.
It takes three input streams and outputs one stream, the first input is the source, and second and third input are displacement maps.
The second input specifies how much to displace pixels along the x-axis, while the third input specifies how much to displace pixels along the y-axis. If one of displacement map streams terminates, last frame from that displacement map will be used.
Note that once generated, displacements maps can be reused over and over again.
A description of the accepted options follows.
Set displace behavior for pixels that are out of range.
Available values are:
Missing pixels are replaced by black pixels.
Adjacent pixels will spread out to replace missing pixels.
Out of range pixels are wrapped so they point to pixels of other side.
Out of range pixels will be replaced with mirrored pixels.
Default is ‘smear’.
ffmpeg -i INPUT -f lavfi -i nullsrc=s=hd720,lutrgb=128:128:128 -f lavfi -i nullsrc=s=hd720,geq='r=128+30*sin(2*PI*X/400+T):g=128+30*sin(2*PI*X/400+T):b=128+30*sin(2*PI*X/400+T)' -lavfi '[0][1][2]displace' OUTPUT
ffmpeg -i INPUT -f lavfi -i nullsrc=hd720,geq='r=128+80*(sin(sqrt((X-W/2)*(X-W/2)+(Y-H/2)*(Y-H/2))/220*2*PI+T)):g=128+80*(sin(sqrt((X-W/2)*(X-W/2)+(Y-H/2)*(Y-H/2))/220*2*PI+T)):b=128+80*(sin(sqrt((X-W/2)*(X-W/2)+(Y-H/2)*(Y-H/2))/220*2*PI+T))' -lavfi '[1]split[x][y],[0][x][y]displace' OUTPUT
Do classification with deep neural networks based on bounding boxes.
The filter accepts the following options:
Specify which DNN backend to use for model loading and execution. This option accepts only openvino now, tensorflow backends will be added.
Set path to model file specifying network architecture and its parameters. Note that different backends use different file formats.
Set the input name of the dnn network.
Set the output name of the dnn network.
Set the confidence threshold (default: 0.5).
Set path to label file specifying the mapping between label id and name. Each label name is written in one line, tailing spaces and empty lines are skipped. The first line is the name of label id 0, and the second line is the name of label id 1, etc. The label id is considered as name if the label file is not provided.
Set the configs to be passed into backend
For tensorflow backend, you can set its configs with sess_config options, please use tools/python/tf_sess_config.py to get the configs for your system.
Do object detection with deep neural networks.
The filter accepts the following options:
Specify which DNN backend to use for model loading and execution. This option accepts only openvino now, tensorflow backends will be added.
Set path to model file specifying network architecture and its parameters. Note that different backends use different file formats.
Set the input name of the dnn network.
Set the output name of the dnn network.
Set the confidence threshold (default: 0.5).
Set path to label file specifying the mapping between label id and name. Each label name is written in one line, tailing spaces and empty lines are skipped. The first line is the name of label id 0 (usually it is ’background’), and the second line is the name of label id 1, etc. The label id is considered as name if the label file is not provided.
Set the configs to be passed into backend. To use async execution, set async (default: set). Roll back to sync execution if the backend does not support async.
Do image processing with deep neural networks. It works together with another filter which converts the pixel format of the Frame to what the dnn network requires.
The filter accepts the following options:
Specify which DNN backend to use for model loading and execution. This option accepts the following values:
TensorFlow backend. To enable this backend you
need to install the TensorFlow for C library (see
https://www.tensorflow.org/install/lang_c) and configure FFmpeg with
--enable-libtensorflow
OpenVINO backend. To enable this backend you
need to build and install the OpenVINO for C library (see
https://github.com/openvinotoolkit/openvino/blob/master/build-instruction.md) and configure FFmpeg with
--enable-libopenvino
(–extra-cflags=-I... –extra-ldflags=-L... might
be needed if the header files and libraries are not installed into system path)
Set path to model file specifying network architecture and its parameters. Note that different backends use different file formats. TensorFlow, OpenVINO backend can load files for only its format.
Set the input name of the dnn network.
Set the output name of the dnn network.
Set the configs to be passed into backend. To use async execution, set async (default: set). Roll back to sync execution if the backend does not support async.
For tensorflow backend, you can set its configs with sess_config options, please use tools/python/tf_sess_config.py to get the configs of TensorFlow backend for your system.
./ffmpeg -i rain.jpg -vf format=rgb24,dnn_processing=dnn_backend=tensorflow:model=can.pb:input=x:output=y derain.jpg
./ffmpeg -i 480p.jpg -vf format=yuv420p,scale=w=iw*2:h=ih*2,dnn_processing=dnn_backend=tensorflow:model=srcnn.pb:input=x:output=y -y srcnn.jpg
./ffmpeg -i 480p.jpg -vf format=yuv420p,dnn_processing=dnn_backend=tensorflow:model=espcn.pb:input=x:output=y:backend_configs=sess_config=0x10022805320e09cdccccccccccec3f20012a01303801 -y tmp.espcn.jpg
Draw a colored box on the input image.
It accepts the following parameters:
The expressions which specify the top left corner coordinates of the box. It defaults to 0.
The expressions which specify the width and height of the box; if 0 they are interpreted as the input width and height. It defaults to 0.
Specify the color of the box to write. For the general syntax of this option,
check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual. If the special
value invert
is used, the box edge color is the same as the
video with inverted luma.
The expression which sets the thickness of the box edge.
A value of fill
will create a filled box. Default value is 3
.
See below for the list of accepted constants.
Applicable if the input has alpha. With value 1
, the pixels of the painted box
will overwrite the video’s color and alpha pixels.
Default is 0
, which composites the box onto the input, leaving the video’s alpha intact.
The parameters for x, y, w and h and t are expressions containing the following constants:
The input display aspect ratio, it is the same as (w / h) * sar.
horizontal and vertical chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
The input width and height.
The input sample aspect ratio.
The x and y offset coordinates where the box is drawn.
The width and height of the drawn box.
Box source can be set as side_data_detection_bboxes if you want to use box data in detection bboxes of side data.
If box_source is set, the x, y, width and height will be ignored and still use box data in detection bboxes of side data. So please do not use this parameter if you were not sure about the box source.
The thickness of the drawn box.
These constants allow the x, y, w, h and t expressions to refer to
each other, so you may for example specify y=x/dar
or h=w/dar
.
drawbox
drawbox=10:20:200:60:red@0.5
The previous example can be specified as:
drawbox=x=10:y=20:w=200:h=60:color=red@0.5
drawbox=x=10:y=10:w=100:h=100:color=pink@0.5:t=fill
drawbox=x=-t:y=0.5*(ih-iw/2.4)-t:w=iw+t*2:h=iw/2.4+t*2:t=2:c=red
Draw a graph using input video metadata.
It accepts the following parameters:
Set 1st frame metadata key from which metadata values will be used to draw a graph.
Set 1st foreground color expression.
Set 2nd frame metadata key from which metadata values will be used to draw a graph.
Set 2nd foreground color expression.
Set 3rd frame metadata key from which metadata values will be used to draw a graph.
Set 3rd foreground color expression.
Set 4th frame metadata key from which metadata values will be used to draw a graph.
Set 4th foreground color expression.
Set minimal value of metadata value.
Set maximal value of metadata value.
Set graph background color. Default is white.
Set graph mode.
Available values for mode is:
Default is line
.
Set slide mode.
Available values for slide is:
Draw new frame when right border is reached.
Replace old columns with new ones.
Scroll from right to left.
Scroll from left to right.
Draw single picture.
Default is frame
.
Set size of graph video. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
The default value is 900x256
.
Set the output frame rate. Default value is 25
.
The foreground color expressions can use the following variables:
Minimal value of metadata value.
Maximal value of metadata value.
Current metadata key value.
The color is defined as 0xAABBGGRR.
Example using metadata from signalstats filter:
signalstats,drawgraph=lavfi.signalstats.YAVG:min=0:max=255
Example using metadata from ebur128 filter:
ebur128=metadata=1,adrawgraph=lavfi.r128.M:min=-120:max=5
Draw a grid on the input image.
It accepts the following parameters:
The expressions which specify the coordinates of some point of grid intersection (meant to configure offset). Both default to 0.
The expressions which specify the width and height of the grid cell, if 0 they are interpreted as the
input width and height, respectively, minus thickness
, so image gets
framed. Default to 0.
Specify the color of the grid. For the general syntax of this option,
check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual. If the special
value invert
is used, the grid color is the same as the
video with inverted luma.
The expression which sets the thickness of the grid line. Default value is 1
.
See below for the list of accepted constants.
Applicable if the input has alpha. With 1
the pixels of the painted grid
will overwrite the video’s color and alpha pixels.
Default is 0
, which composites the grid onto the input, leaving the video’s alpha intact.
The parameters for x, y, w and h and t are expressions containing the following constants:
The input display aspect ratio, it is the same as (w / h) * sar.
horizontal and vertical chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
The input grid cell width and height.
The input sample aspect ratio.
The x and y coordinates of some point of grid intersection (meant to configure offset).
The width and height of the drawn cell.
The thickness of the drawn cell.
These constants allow the x, y, w, h and t expressions to refer to
each other, so you may for example specify y=x/dar
or h=w/dar
.
drawgrid=width=100:height=100:thickness=2:color=red@0.5
drawgrid=w=iw/3:h=ih/3:t=2:c=white@0.5
Draw a text string or text from a specified file on top of a video, using the libfreetype library.
To enable compilation of this filter, you need to configure FFmpeg with
--enable-libfreetype
and --enable-libharfbuzz
.
To enable default font fallback and the font option you need to
configure FFmpeg with --enable-libfontconfig
.
To enable the text_shaping option, you need to configure FFmpeg with
--enable-libfribidi
.
It accepts the following parameters:
Used to draw a box around text using the background color. The value must be either 1 (enable) or 0 (disable). The default value of box is 0.
Set the width of the border to be drawn around the box using boxcolor. The value must be specified using one of the following formats:
boxborderw=10
set the width of all the borders to 10
boxborderw=10|20
set the width of the top and bottom borders to 10
and the width of the left and right borders to 20
boxborderw=10|20|30
set the width of the top border to 10, the width
of the bottom border to 30 and the width of the left and right borders to 20
boxborderw=10|20|30|40
set the borders width to 10 (top), 20 (right),
30 (bottom), 40 (left)
The default value of boxborderw is "0".
The color to be used for drawing box around text. For the syntax of this option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual.
The default value of boxcolor is "white".
Set the line spacing in pixels. The default value of line_spacing is 0.
Set the vertical and horizontal alignment of the text with respect to the box boundaries. The value is combination of flags, one for the vertical alignment (T=top, M=middle, B=bottom) and one for the horizontal alignment (L=left, C=center, R=right). Please note that tab characters are only supported with the left horizontal alignment.
Specify what the y value is referred to. Possible values are:
text
the top of the highest glyph of the first text line is placed at y
baseline
the baseline of the first text line is placed at y
font
the baseline of the first text line is placed at y plus the
ascent (in pixels) defined in the font metrics
The default value of y_align is "text" for backward compatibility.
Set the width of the border to be drawn around the text using bordercolor. The default value of borderw is 0.
Set the color to be used for drawing border around text. For the syntax of this option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual.
The default value of bordercolor is "black".
Select how the text is expanded. Can be either none
,
strftime
(deprecated) or normal
(default). See the
Text expansion section below for details.
Set a start time for the count. Value is in microseconds. Only applied
in the deprecated strftime
expansion mode. To emulate in normal expansion
mode use the pts
function, supplying the start time (in seconds)
as the second argument.
If true, check and fix text coords to avoid clipping.
The color to be used for drawing fonts. For the syntax of this option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual.
The default value of fontcolor is "black".
String which is expanded the same way as text to obtain dynamic fontcolor value. By default this option has empty value and is not processed. When this option is set, it overrides fontcolor option.
The font family to be used for drawing text. By default Sans.
The font file to be used for drawing text. The path must be included. This parameter is mandatory if the fontconfig support is disabled.
Draw the text applying alpha blending. The value can be a number between 0.0 and 1.0. The expression accepts the same variables x, y as well. The default value is 1. Please see fontcolor_expr.
The font size to be used for drawing text. The default value of fontsize is 16.
If set to 1, attempt to shape the text (for example, reverse the order of right-to-left text and join Arabic characters) before drawing it. Otherwise, just draw the text exactly as given. By default 1 (if supported).
The flags to be used for loading the fonts.
The flags map the corresponding flags supported by libfreetype, and are a combination of the following values:
Default value is "default".
For more information consult the documentation for the FT_LOAD_* libfreetype flags.
The color to be used for drawing a shadow behind the drawn text. For the syntax of this option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual.
The default value of shadowcolor is "black".
Set the width of the box to be drawn around text. The default value of boxw is computed automatically to match the text width
Set the height of the box to be drawn around text. The default value of boxh is computed automatically to match the text height
The x and y offsets for the text shadow position with respect to the position of the text. They can be either positive or negative values. The default value for both is "0".
The starting frame number for the n/frame_num variable. The default value is "0".
The size in number of spaces to use for rendering the tab. Default value is 4.
Set the initial timecode representation in "hh:mm:ss[:;.]ff" format. It can be used with or without text parameter. timecode_rate option must be specified.
Set the timecode frame rate (timecode only). Value will be rounded to nearest integer. Minimum value is "1". Drop-frame timecode is supported for frame rates 30 & 60.
If set to 1, the output of the timecode option will wrap around at 24 hours. Default is 0 (disabled).
The text string to be drawn. The text must be a sequence of UTF-8 encoded characters. This parameter is mandatory if no file is specified with the parameter textfile.
A text file containing text to be drawn. The text must be a sequence of UTF-8 encoded characters.
This parameter is mandatory if no text string is specified with the parameter text.
If both text and textfile are specified, an error is thrown.
Text source should be set as side_data_detection_bboxes if you want to use text data in detection bboxes of side data.
If text source is set, text and textfile will be ignored and still use text data in detection bboxes of side data. So please do not use this parameter if you are not sure about the text source.
The textfile will be reloaded at specified frame interval. Be sure to update textfile atomically, or it may be read partially, or even fail. Range is 0 to INT_MAX. Default is 0.
The expressions which specify the offsets where text will be drawn within the video frame. They are relative to the top/left border of the output image.
The default value of x and y is "0".
See below for the list of accepted constants and functions.
The parameters for x and y are expressions containing the following constants and functions:
input display aspect ratio, it is the same as (w / h) * sar
horizontal and vertical chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
the height of each text line
the input height
the input width
the maximum distance from the baseline to the highest/upper grid coordinate used to place a glyph outline point, for all the rendered glyphs. It is a positive value, due to the grid’s orientation with the Y axis upwards.
the maximum distance from the baseline to the lowest grid coordinate used to place a glyph outline point, for all the rendered glyphs. This is a negative value, due to the grid’s orientation, with the Y axis upwards.
maximum glyph height, that is the maximum height for all the glyphs contained in the rendered text, it is equivalent to ascent - descent.
maximum glyph width, that is the maximum width for all the glyphs contained in the rendered text
the ascent size defined in the font metrics
the descent size defined in the font metrics
the maximum ascender of the glyphs of the first text line
the maximum descender of the glyphs of the last text line
the number of input frame, starting from 0
return a random number included between min and max
The input sample aspect ratio.
timestamp expressed in seconds, NAN if the input timestamp is unknown
the height of the rendered text
the width of the rendered text
the x and y offset coordinates where the text is drawn.
These parameters allow the x and y expressions to refer
to each other, so you can for example specify y=x/dar
.
A one character description of the current frame’s picture type.
The current packet’s position in the input file or stream (in bytes, from the start of the input). A value of -1 indicates this info is not available.
The current packet’s duration, in seconds.
The current packet’s size (in bytes).
If expansion is set to strftime
, the filter recognizes
sequences accepted by the strftime
C function in the provided
text and expands them accordingly. Check the documentation of
strftime
. This feature is deprecated in favor of normal
expansion with the gmtime
or localtime
expansion
functions.
If expansion is set to none
, the text is printed verbatim.
If expansion is set to normal
(which is the default),
the following expansion mechanism is used.
The backslash character ‘\’, followed by any character, always expands to the second character.
Sequences of the form %{...}
are expanded. The text between the
braces is a function name, possibly followed by arguments separated by ’:’.
If the arguments contain special characters or delimiters (’:’ or ’}’),
they should be escaped.
Note that they probably must also be escaped as the value for the text option in the filter argument string and as the filter argument in the filtergraph description, and possibly also for the shell, that makes up to four levels of escaping; using a text file with the textfile option avoids these problems.
The following functions are available:
expr, e
The expression evaluation result.
It must take one argument specifying the expression to be evaluated, which accepts the same constants and functions as the x and y values. Note that not all constants should be used, for example the text size is not known when evaluating the expression, so the constants text_w and text_h will have an undefined value.
expr_int_format, eif
Evaluate the expression’s value and output as formatted integer.
The first argument is the expression to be evaluated, just as for the expr function.
The second argument specifies the output format. Allowed values are ‘x’,
‘X’, ‘d’ and ‘u’. They are treated exactly as in the
printf
function.
The third parameter is optional and sets the number of positions taken by the output.
It can be used to add padding with zeros from the left.
gmtime
The time at which the filter is running, expressed in UTC.
It can accept an argument: a strftime
C function format string.
The format string is extended to support the variable %[1-6]N
which prints fractions of the second with optionally specified number of digits.
localtime
The time at which the filter is running, expressed in the local time zone.
It can accept an argument: a strftime
C function format string.
The format string is extended to support the variable %[1-6]N
which prints fractions of the second with optionally specified number of digits.
metadata
Frame metadata. Takes one or two arguments.
The first argument is mandatory and specifies the metadata key.
The second argument is optional and specifies a default value, used when the metadata key is not found or empty.
Available metadata can be identified by inspecting entries
starting with TAG included within each frame section
printed by running ffprobe -show_frames
.
String metadata generated in filters leading to the drawtext filter are also available.
n, frame_num
The frame number, starting from 0.
pict_type
A one character description of the current picture type.
pts
The timestamp of the current frame. It can take up to three arguments.
The first argument is the format of the timestamp; it defaults to flt
for seconds as a decimal number with microsecond accuracy; hms
stands
for a formatted [-]HH:MM:SS.mmm timestamp with millisecond accuracy.
gmtime
stands for the timestamp of the frame formatted as UTC time;
localtime
stands for the timestamp of the frame formatted as
local time zone time.
The second argument is an offset added to the timestamp.
If the format is set to hms
, a third argument 24HH
may be
supplied to present the hour part of the formatted timestamp in 24h format
(00-23).
If the format is set to localtime
or gmtime
, a third
argument may be supplied: a strftime
C function format string.
By default, YYYY-MM-DD HH:MM:SS format will be used.
This filter supports altering parameters via commands:
Alter existing filter parameters.
Syntax for the argument is the same as for filter invocation, e.g.
fontsize=56:fontcolor=green:text='Hello World'
Full filter invocation with sendcmd would look like this:
sendcmd=c='56.0 drawtext reinit fontsize=56\:fontcolor=green\:text=Hello\\ World'
If the entire argument can’t be parsed or applied as valid values then the filter will continue with its existing parameters.
The following options are also supported as commands:
drawtext="fontfile=/usr/share/fonts/truetype/freefont/FreeSerif.ttf: text='Test Text'"
drawtext="fontfile=/usr/share/fonts/truetype/freefont/FreeSerif.ttf: text='Test Text':\ x=100: y=50: fontsize=24: fontcolor=yellow@0.2: box=1: boxcolor=red@0.2"
Note that the double quotes are not necessary if spaces are not used within the parameter list.
drawtext="fontsize=30:fontfile=FreeSerif.ttf:text='hello world':x=(w-text_w)/2:y=(h-text_h)/2"
drawtext="fontsize=30:fontfile=FreeSerif.ttf:text='hello world':x=if(eq(mod(t\,30)\,0)\,rand(0\,(w-text_w))\,x):y=if(eq(mod(t\,30)\,0)\,rand(0\,(h-text_h))\,y)"
drawtext="fontsize=15:fontfile=FreeSerif.ttf:text=LONG_LINE:y=h-line_h:x=-50*t"
drawtext="fontsize=20:fontfile=FreeSerif.ttf:textfile=CREDITS:y=h-20*t"
drawtext="fontsize=60:fontfile=FreeSerif.ttf:fontcolor=green:text=g:x=(w-max_glyph_w)/2:y=h/2-ascent"
drawtext="fontfile=FreeSerif.ttf:fontcolor=white:x=100:y=x/dar:enable=lt(mod(t\,3)\,1):text='blink'"
drawtext='fontfile=Linux Libertine O-40\:style=Semibold:text=FFmpeg'
drawtext="text='Test Text': fontsize=h/30: x=(w-text_w)/2: y=(h-text_h*2)"
strftime
C function):
drawtext='fontfile=FreeSans.ttf:text=%{localtime\:%a %b %d %Y}'
#!/bin/sh DS=1.0 # display start DE=10.0 # display end FID=1.5 # fade in duration FOD=5 # fade out duration ffplay -f lavfi "color,drawtext=text=TEST:fontsize=50:fontfile=FreeSerif.ttf:fontcolor_expr=ff0000%{eif\\\\: clip(255*(1*between(t\\, $DS + $FID\\, $DE - $FOD) + ((t - $DS)/$FID)*between(t\\, $DS\\, $DS + $FID) + (-(t - $DE)/$FOD)*between(t\\, $DE - $FOD\\, $DE) )\\, 0\\, 255) \\\\: x\\\\: 2 }"
drawtext=fontfile=FreeSans.ttf:text=DOG:fontsize=24:x=10:y=20+24-max_glyph_a, drawtext=fontfile=FreeSans.ttf:text=cow:fontsize=24:x=80:y=20+24-max_glyph_a
drawtext="fontsize=20:fontcolor=white:fontfile=FreeSans.ttf:text='%{metadata\:lavf.image2dec.source_basename\:NA}':x=10:y=10"
For more information about libfreetype, check: http://www.freetype.org/.
For more information about fontconfig, check: http://freedesktop.org/software/fontconfig/fontconfig-user.html.
For more information about libfribidi, check: http://fribidi.org/.
For more information about libharfbuzz, check: https://github.com/harfbuzz/harfbuzz.
Detect and draw edges. The filter uses the Canny Edge Detection algorithm.
The filter accepts the following options:
Set low and high threshold values used by the Canny thresholding algorithm.
The high threshold selects the "strong" edge pixels, which are then connected through 8-connectivity with the "weak" edge pixels selected by the low threshold.
low and high threshold values must be chosen in the range [0,1], and low should be lesser or equal to high.
Default value for low is 20/255
, and default value for high
is 50/255
.
Define the drawing mode.
Draw white/gray wires on black background.
Mix the colors to create a paint/cartoon effect.
Apply Canny edge detector on all selected planes.
Default value is wires.
Select planes for filtering. By default all available planes are filtered.
edgedetect=low=0.1:high=0.4
edgedetect=mode=colormix:high=0
Apply a posterize effect using the ELBG (Enhanced LBG) algorithm.
For each input image, the filter will compute the optimal mapping from the input to the output given the codebook length, that is the number of distinct output colors.
This filter accepts the following options.
Set codebook length. The value must be a positive integer, and represents the number of distinct output colors. Default value is 256.
Set the maximum number of iterations to apply for computing the optimal mapping. The higher the value the better the result and the higher the computation time. Default value is 1.
Set a random seed, must be an integer included between 0 and UINT32_MAX. If not specified, or if explicitly set to -1, the filter will try to use a good random seed on a best effort basis.
Set pal8 output pixel format. This option does not work with codebook length greater than 256. Default is disabled.
Include alpha values in the quantization calculation. Allows creating palettized output images (e.g. PNG8) with multiple alpha smooth blending.
Measure graylevel entropy in histogram of color channels of video frames.
It accepts the following parameters:
Can be either normal or diff. Default is normal.
diff mode measures entropy of histogram delta values, absolute differences between neighbour histogram values.
Apply the EPX magnification filter which is designed for pixel art.
It accepts the following option:
Set the scaling dimension: 2
for 2xEPX
, 3
for
3xEPX
.
Default is 3
.
Set brightness, contrast, saturation and approximate gamma adjustment.
The filter accepts the following options:
Set the contrast expression. The value must be a float value in range
-1000.0
to 1000.0
. The default value is "1".
Set the brightness expression. The value must be a float value in
range -1.0
to 1.0
. The default value is "0".
Set the saturation expression. The value must be a float in
range 0.0
to 3.0
. The default value is "1".
Set the gamma expression. The value must be a float in range
0.1
to 10.0
. The default value is "1".
Set the gamma expression for red. The value must be a float in
range 0.1
to 10.0
. The default value is "1".
Set the gamma expression for green. The value must be a float in range
0.1
to 10.0
. The default value is "1".
Set the gamma expression for blue. The value must be a float in range
0.1
to 10.0
. The default value is "1".
Set the gamma weight expression. It can be used to reduce the effect
of a high gamma value on bright image areas, e.g. keep them from
getting overamplified and just plain white. The value must be a float
in range 0.0
to 1.0
. A value of 0.0
turns the
gamma correction all the way down while 1.0
leaves it at its
full strength. Default is "1".
Set when the expressions for brightness, contrast, saturation and gamma expressions are evaluated.
It accepts the following values:
only evaluate expressions once during the filter initialization or when a command is processed
evaluate expressions for each incoming frame
Default value is ‘init’.
The expressions accept the following parameters:
frame count of the input frame starting from 0
byte position of the corresponding packet in the input file, NAN if unspecified; deprecated, do not use
frame rate of the input video, NAN if the input frame rate is unknown
timestamp expressed in seconds, NAN if the input timestamp is unknown
The filter supports the following commands:
Set the contrast expression.
Set the brightness expression.
Set the saturation expression.
Set the gamma expression.
Set the gamma_r expression.
Set gamma_g expression.
Set gamma_b expression.
Set gamma_weight expression.
The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Apply erosion effect to the video.
This filter replaces the pixel by the local(3x3) minimum.
It accepts the following options:
Limit the maximum change for each plane, default is 65535. If 0, plane will remain unchanged.
Flag which specifies the pixel to refer to. Default is 255 i.e. all eight pixels are used.
Flags to local 3x3 coordinates maps like this:
1 2 3 4 5 6 7 8
This filter supports the all above options as commands.
Deinterlace the input video ("estdif" stands for "Edge Slope Tracing Deinterlacing Filter").
Spatial only filter that uses edge slope tracing algorithm to interpolate missing lines. It accepts the following parameters:
The interlacing mode to adopt. It accepts one of the following values:
Output one frame for each frame.
Output one frame for each field.
The default value is field
.
The picture field parity assumed for the input interlaced video. It accepts one of the following values:
Assume the top field is first.
Assume the bottom field is first.
Enable automatic detection of field parity.
The default value is auto
.
If the interlacing is unknown or the decoder does not export this information,
top field first will be assumed.
Specify which frames to deinterlace. Accepts one of the following values:
Deinterlace all frames.
Only deinterlace frames marked as interlaced.
The default value is all
.
Specify the search radius for edge slope tracing. Default value is 1. Allowed range is from 1 to 15.
Specify the search radius for best edge matching. Default value is 2. Allowed range is from 0 to 15.
Specify the edge cost for edge matching. Default value is 2. Allowed range is from 0 to 50.
Specify the middle cost for edge matching. Default value is 1. Allowed range is from 0 to 50.
Specify the distance cost for edge matching. Default value is 1. Allowed range is from 0 to 50.
Specify the interpolation used. Default is 4-point interpolation. It accepts one of the following values:
Two-point interpolation.
Four-point interpolation.
Six-point interpolation.
This filter supports same commands as options.
Adjust exposure of the video stream.
The filter accepts the following options:
Set the exposure correction in EV. Allowed range is from -3.0 to 3.0 EV Default value is 0 EV.
Set the black level correction. Allowed range is from -1.0 to 1.0. Default value is 0.
This filter supports same commands as options.
Extract color channel components from input video stream into separate grayscale video streams.
The filter accepts the following option:
Set plane(s) to extract.
Available values for planes are:
Choosing planes not available in the input will result in an error.
That means you cannot select r
, g
, b
planes
with y
, u
, v
planes at same time.
ffmpeg -i video.avi -filter_complex 'extractplanes=y+u+v[y][u][v]' -map '[y]' y.avi -map '[u]' u.avi -map '[v]' v.avi
Apply a fade-in/out effect to the input video.
It accepts the following parameters:
The effect type can be either "in" for a fade-in, or "out" for a fade-out
effect.
Default is in
.
Specify the number of the frame to start applying the fade effect at. Default is 0.
The number of frames that the fade effect lasts. At the end of the fade-in effect, the output video will have the same intensity as the input video. At the end of the fade-out transition, the output video will be filled with the selected color. Default is 25.
If set to 1, fade only alpha channel, if one exists on the input. Default value is 0.
Specify the timestamp (in seconds) of the frame to start to apply the fade effect. If both start_frame and start_time are specified, the fade will start at whichever comes last. Default is 0.
The number of seconds for which the fade effect has to last. At the end of the fade-in effect the output video will have the same intensity as the input video, at the end of the fade-out transition the output video will be filled with the selected color. If both duration and nb_frames are specified, duration is used. Default is 0 (nb_frames is used by default).
Specify the color of the fade. Default is "black".
fade=in:0:30
The command above is equivalent to:
fade=t=in:s=0:n=30
fade=out:155:45 fade=type=out:start_frame=155:nb_frames=45
fade=in:0:25, fade=out:975:25
fade=in:5:20:color=yellow
fade=in:0:25:alpha=1
fade=t=in:st=5.5:d=0.5
Apply feedback video filter.
This filter pass cropped input frames to 2nd output. From there it can be filtered with other video filters. After filter receives frame from 2nd input, that frame is combined on top of original frame from 1st input and passed to 1st output.
The typical usage is filter only part of frame.
The filter accepts the following options:
Set the top left crop position.
Set the crop size.
[in][blurin]feedback=x=0:y=0:w=100:h=100[out][blurout];[blurout]gblur=8[blurin]
[in][blurin]feedback=x=0:y=0:w=100:h=100[out][blurout];[blurout]drawbox=x=0:y=0:w=100:h=100:t=100[blurin]
[in][blurin]feedback=x=320:y=240:w=100:h=100[out][blurout];[blurout]pixelize[blurin]
Denoise frames using 3D FFT (frequency domain filtering).
The filter accepts the following options:
Set the noise sigma constant. This sets denoising strength. Default value is 1. Allowed range is from 0 to 30. Using very high sigma with low overlap may give blocking artifacts.
Set amount of denoising. By default all detected noise is reduced. Default value is 1. Allowed range is from 0 to 1.
Set size of block in pixels, Default is 32, can be 8 to 256.
Set block overlap. Default is 0.5. Allowed range is from 0.2 to 0.8.
Set denoising method. Default is wiener
, can also be hard
.
Set number of previous frames to use for denoising. By default is set to 0.
Set number of next frames to to use for denoising. By default is set to 0.
Set planes which will be filtered, by default are all available filtered except alpha.
Apply arbitrary expressions to samples in frequency domain
Adjust the dc value (gain) of the luma plane of the image. The filter
accepts an integer value in range 0
to 1000
. The default
value is set to 0
.
Adjust the dc value (gain) of the 1st chroma plane of the image. The
filter accepts an integer value in range 0
to 1000
. The
default value is set to 0
.
Adjust the dc value (gain) of the 2nd chroma plane of the image. The
filter accepts an integer value in range 0
to 1000
. The
default value is set to 0
.
Set the frequency domain weight expression for the luma plane.
Set the frequency domain weight expression for the 1st chroma plane.
Set the frequency domain weight expression for the 2nd chroma plane.
Set when the expressions are evaluated.
It accepts the following values:
Only evaluate expressions once during the filter initialization.
Evaluate expressions for each incoming frame.
Default value is ‘init’.
The filter accepts the following variables:
The coordinates of the current sample.
The width and height of the image.
The number of input frame, starting from 0.
The size of FFT array for horizontal and vertical processing.
fftfilt=dc_Y=128:weight_Y='squish(1-(Y+X)/100)'
fftfilt=dc_Y=0:weight_Y='squish((Y+X)/100-1)'
fftfilt=dc_Y=0:weight_Y='1+squish(1-(Y+X)/100)'
fftfilt=dc_Y=0:weight_Y='exp(-4 * ((Y+X)/(W+H)))'
Extract a single field from an interlaced image using stride arithmetic to avoid wasting CPU time. The output frames are marked as non-interlaced.
The filter accepts the following options:
Specify whether to extract the top (if the value is 0
or
top
) or the bottom field (if the value is 1
or
bottom
).
Create new frames by copying the top and bottom fields from surrounding frames supplied as numbers by the hint file.
Set file containing hints: absolute/relative frame numbers.
There must be one line for each frame in a clip. Each line must contain two
numbers separated by the comma, optionally followed by -
or +
.
Numbers supplied on each line of file can not be out of [N-1,N+1] where N
is current frame number for absolute
mode or out of [-1, 1] range
for relative
mode. First number tells from which frame to pick up top
field and second number tells from which frame to pick up bottom field.
If optionally followed by +
output frame will be marked as interlaced,
else if followed by -
output frame will be marked as progressive, else
it will be marked same as input frame.
If optionally followed by t
output frame will use only top field, or in
case of b
it will use only bottom field.
If line starts with #
or ;
that line is skipped.
Can be item absolute
or relative
or pattern
. Default is absolute
.
The pattern
mode is same as relative
mode, except at last entry of file if there
are more frames to process than hint
file is seek back to start.
Example of first several lines of hint
file for relative
mode:
0,0 - # first frame 1,0 - # second frame, use third's frame top field and second's frame bottom field 1,0 - # third frame, use fourth's frame top field and third's frame bottom field 1,0 - 0,0 - 0,0 - 1,0 - 1,0 - 1,0 - 0,0 - 0,0 - 1,0 - 1,0 - 1,0 - 0,0 -
Field matching filter for inverse telecine. It is meant to reconstruct the
progressive frames from a telecined stream. The filter does not drop duplicated
frames, so to achieve a complete inverse telecine fieldmatch
needs to be
followed by a decimation filter such as decimate in the filtergraph.
The separation of the field matching and the decimation is notably motivated by
the possibility of inserting a de-interlacing filter fallback between the two.
If the source has mixed telecined and real interlaced content,
fieldmatch
will not be able to match fields for the interlaced parts.
But these remaining combed frames will be marked as interlaced, and thus can be
de-interlaced by a later filter such as yadif before decimation.
In addition to the various configuration options, fieldmatch
can take an
optional second stream, activated through the ppsrc option. If
enabled, the frames reconstruction will be based on the fields and frames from
this second stream. This allows the first input to be pre-processed in order to
help the various algorithms of the filter, while keeping the output lossless
(assuming the fields are matched properly). Typically, a field-aware denoiser,
or brightness/contrast adjustments can help.
Note that this filter uses the same algorithms as TIVTC/TFM (AviSynth project)
and VIVTC/VFM (VapourSynth project). The later is a light clone of TFM from
which fieldmatch
is based on. While the semantic and usage are very
close, some behaviour and options names can differ.
The decimate filter currently only works for constant frame rate input.
If your input has mixed telecined (30fps) and progressive content with a lower
framerate like 24fps use the following filterchain to produce the necessary cfr
stream: dejudder,fps=30000/1001,fieldmatch,decimate
.
The filter accepts the following options:
Specify the assumed field order of the input stream. Available values are:
Auto detect parity (use FFmpeg’s internal parity value).
Assume bottom field first.
Assume top field first.
Note that it is sometimes recommended not to trust the parity announced by the stream.
Default value is auto.
Set the matching mode or strategy to use. pc mode is the safest in the sense that it won’t risk creating jerkiness due to duplicate frames when possible, but if there are bad edits or blended fields it will end up outputting combed frames when a good match might actually exist. On the other hand, pcn_ub mode is the most risky in terms of creating jerkiness, but will almost always find a good frame if there is one. The other values are all somewhere in between pc and pcn_ub in terms of risking jerkiness and creating duplicate frames versus finding good matches in sections with bad edits, orphaned fields, blended fields, etc.
More details about p/c/n/u/b are available in p/c/n/u/b meaning section.
Available values are:
2-way matching (p/c)
2-way matching, and trying 3rd match if still combed (p/c + n)
2-way matching, and trying 3rd match (same order) if still combed (p/c + u)
2-way matching, trying 3rd match if still combed, and trying 4th/5th matches if still combed (p/c + n + u/b)
3-way matching (p/c/n)
3-way matching, and trying 4th/5th matches if all 3 of the original matches are detected as combed (p/c/n + u/b)
The parenthesis at the end indicate the matches that would be used for that mode assuming order=tff (and field on auto or top).
In terms of speed pc mode is by far the fastest and pcn_ub is the slowest.
Default value is pc_n.
Mark the main input stream as a pre-processed input, and enable the secondary input stream as the clean source to pick the fields from. See the filter introduction for more details. It is similar to the clip2 feature from VFM/TFM.
Default value is 0
(disabled).
Set the field to match from. It is recommended to set this to the same value as order unless you experience matching failures with that setting. In certain circumstances changing the field that is used to match from can have a large impact on matching performance. Available values are:
Automatic (same value as order).
Match from the bottom field.
Match from the top field.
Default value is auto.
Set whether or not chroma is included during the match comparisons. In most
cases it is recommended to leave this enabled. You should set this to 0
only if your clip has bad chroma problems such as heavy rainbowing or other
artifacts. Setting this to 0
could also be used to speed things up at
the cost of some accuracy.
Default value is 1
.
These define an exclusion band which excludes the lines between y0 and
y1 from being included in the field matching decision. An exclusion
band can be used to ignore subtitles, a logo, or other things that may
interfere with the matching. y0 sets the starting scan line and
y1 sets the ending line; all lines in between y0 and
y1 (including y0 and y1) will be ignored. Setting
y0 and y1 to the same value will disable the feature.
y0 and y1 defaults to 0
.
Set the scene change detection threshold as a percentage of maximum change on
the luma plane. Good values are in the [8.0, 14.0]
range. Scene change
detection is only relevant in case combmatch=sc. The range for
scthresh is [0.0, 100.0]
.
Default value is 12.0
.
When combatch is not none, fieldmatch
will take into
account the combed scores of matches when deciding what match to use as the
final match. Available values are:
No final matching based on combed scores.
Combed scores are only used when a scene change is detected.
Use combed scores all the time.
Default is sc.
Force fieldmatch
to calculate the combed metrics for certain matches and
print them. This setting is known as micout in TFM/VFM vocabulary.
Available values are:
No forced calculation.
Force p/c/n calculations.
Force p/c/n/u/b calculations.
Default value is none.
This is the area combing threshold used for combed frame detection. This
essentially controls how "strong" or "visible" combing must be to be detected.
Larger values mean combing must be more visible and smaller values mean combing
can be less visible or strong and still be detected. Valid settings are from
-1
(every pixel will be detected as combed) to 255
(no pixel will
be detected as combed). This is basically a pixel difference value. A good
range is [8, 12]
.
Default value is 9
.
Sets whether or not chroma is considered in the combed frame decision. Only disable this if your source has chroma problems (rainbowing, etc.) that are causing problems for the combed frame detection with chroma enabled. Actually, using chroma=0 is usually more reliable, except for the case where there is chroma only combing in the source.
Default value is 0
.
Respectively set the x-axis and y-axis size of the window used during combed frame detection. This has to do with the size of the area in which combpel pixels are required to be detected as combed for a frame to be declared combed. See the combpel parameter description for more info. Possible values are any number that is a power of 2 starting at 4 and going up to 512.
Default value is 16
.
The number of combed pixels inside any of the blocky by
blockx size blocks on the frame for the frame to be detected as
combed. While cthresh controls how "visible" the combing must be, this
setting controls "how much" combing there must be in any localized area (a
window defined by the blockx and blocky settings) on the
frame. Minimum value is 0
and maximum is blocky x blockx
(at
which point no frames will ever be detected as combed). This setting is known
as MI in TFM/VFM vocabulary.
Default value is 80
.
We assume the following telecined stream:
Top fields: 1 2 2 3 4 Bottom fields: 1 2 3 4 4
The numbers correspond to the progressive frame the fields relate to. Here, the first two frames are progressive, the 3rd and 4th are combed, and so on.
When fieldmatch
is configured to run a matching from bottom
(field=bottom) this is how this input stream get transformed:
Input stream: T 1 2 2 3 4 B 1 2 3 4 4 <-- matching reference Matches: c c n n c Output stream: T 1 2 3 4 4 B 1 2 3 4 4
As a result of the field matching, we can see that some frames get duplicated. To perform a complete inverse telecine, you need to rely on a decimation filter after this operation. See for instance the decimate filter.
The same operation now matching from top fields (field=top) looks like this:
Input stream: T 1 2 2 3 4 <-- matching reference B 1 2 3 4 4 Matches: c c p p c Output stream: T 1 2 2 3 4 B 1 2 2 3 4
In these examples, we can see what p, c and n mean; basically, they refer to the frame and field of the opposite parity:
The u and b matching are a bit special in the sense that they match from the opposite parity flag. In the following examples, we assume that we are currently matching the 2nd frame (Top:2, bottom:2). According to the match, a ’x’ is placed above and below each matched fields.
With bottom matching (field=bottom):
Match: c p n b u x x x x x Top 1 2 2 1 2 2 1 2 2 1 2 2 1 2 2 Bottom 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 x x x x x Output frames: 2 1 2 2 2 2 2 2 1 3
With top matching (field=top):
Match: c p n b u x x x x x Top 1 2 2 1 2 2 1 2 2 1 2 2 1 2 2 Bottom 1 2 3 1 2 3 1 2 3 1 2 3 1 2 3 x x x x x Output frames: 2 2 2 1 2 2 1 3 2 2
Simple IVTC of a top field first telecined stream:
fieldmatch=order=tff:combmatch=none, decimate
Advanced IVTC, with fallback on yadif for still combed frames:
fieldmatch=order=tff:combmatch=full, yadif=deint=interlaced, decimate
Transform the field order of the input video.
It accepts the following parameters:
The output field order. Valid values are tff for top field first or bff for bottom field first.
The default value is ‘tff’.
The transformation is done by shifting the picture content up or down by one line, and filling the remaining line with appropriate picture content. This method is consistent with most broadcast field order converters.
If the input video is not flagged as being interlaced, or it is already flagged as being of the required output field order, then this filter does not alter the incoming video.
It is very useful when converting to or from PAL DV material, which is bottom field first.
For example:
ffmpeg -i in.vob -vf "fieldorder=bff" out.dv
Buffer input images and send them when they are requested.
It is mainly useful when auto-inserted by the libavfilter framework.
It does not take parameters.
Fill borders of the input video, without changing video stream dimensions. Sometimes video can have garbage at the four edges and you may not want to crop video input to keep size multiple of some number.
This filter accepts the following options:
Number of pixels to fill from left border.
Number of pixels to fill from right border.
Number of pixels to fill from top border.
Number of pixels to fill from bottom border.
Set fill mode.
It accepts the following values:
fill pixels using outermost pixels
fill pixels using mirroring (half sample symmetric)
fill pixels with constant value
fill pixels using reflecting (whole sample symmetric)
fill pixels using wrapping
fade pixels to constant value
fill pixels at top and bottom with weighted averages pixels near borders
Default is smear.
Set color for pixels in fixed or fade mode. Default is black.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Find a rectangular object in the input video.
The object to search for must be specified as a gray8 image specified with the object option.
For each possible match, a score is computed. If the score reaches the specified threshold, the object is considered found.
If the input video contains multiple instances of the object, the filter will find only one of them.
When an object is found, the following metadata entries are set in the matching frame:
width of object
height of object
x position of object
y position of object
match score of the found object
It accepts the following options:
Filepath of the object image, needs to be in gray8.
Detection threshold, expressed as a decimal number in the range 0-1.
A threshold value of 0.01 means only exact matches, a threshold of 0.99 means almost everything matches.
Default value is 0.5.
Number of mipmaps, default is 3.
Specifies the rectangle in which to search.
Discard frames where object is not detected. Default is disabled.
ffmpeg
:
ffmpeg -i file.ts -vf find_rect=newref.pgm,cover_rect=cover.jpg:mode=cover new.mkv
ffprobe
and write
it to a log file:
ffprobe -f lavfi movie=test.mp4,find_rect=object=object.pgm:threshold=0.3 \ -show_entries frame=pkt_pts_time:frame_tags=lavfi.rect.x,lavfi.rect.y \ -of csv -o find_rect.csv
Flood area with values of same pixel components with another values.
It accepts the following options:
Set pixel x coordinate.
Set pixel y coordinate.
Set source #0 component value.
Set source #1 component value.
Set source #2 component value.
Set source #3 component value.
Set destination #0 component value.
Set destination #1 component value.
Set destination #2 component value.
Set destination #3 component value.
Convert the input video to one of the specified pixel formats. Libavfilter will try to pick one that is suitable as input to the next filter.
It accepts the following parameters:
A ’|’-separated list of pixel format names, such as "pix_fmts=yuv420p|monow|rgb24".
Convert the video to specified constant frame rate by duplicating or dropping frames as necessary.
It accepts the following parameters:
The desired output frame rate. It accepts expressions containing the following constants:
The input’s frame rate
NTSC frame rate of 30000/1001
PAL frame rate of 25.0
Film frame rate of 24.0
NTSC-film frame rate of 24000/1001
The default is 25
.
Assume the first PTS should be the given value, in seconds. This allows for padding/trimming at the start of stream. By default, no assumption is made about the first frame’s expected PTS, so no padding or trimming is done. For example, this could be set to 0 to pad the beginning with duplicates of the first frame if a video stream starts after the audio stream or to trim any frames with a negative PTS.
Timestamp (PTS) rounding method.
Possible values are:
round towards 0
round away from 0
round towards -infinity
round towards +infinity
round to nearest
The default is near
.
Action performed when reading the last frame.
Possible values are:
Use same timestamp rounding method as used for other frames.
Pass through last frame if input duration has not been reached yet.
The default is round
.
Alternatively, the options can be specified as a flat string: fps[:start_time[:round]].
See also the setpts filter.
fps=fps=25
fps=fps=film:round=near
Pack two different video streams into a stereoscopic video, setting proper metadata on supported codecs. The two views should have the same size and framerate and processing will stop when the shorter video ends. Please note that you may conveniently adjust view properties with the scale and fps filters.
It accepts the following parameters:
The desired packing format. Supported values are:
The views are next to each other (default).
The views are on top of each other.
The views are packed by line.
The views are packed by column.
The views are temporally interleaved.
Some examples:
# Convert left and right views into a frame-sequential video ffmpeg -i LEFT -i RIGHT -filter_complex framepack=frameseq OUTPUT # Convert views into a side-by-side video with the same output resolution as the input ffmpeg -i LEFT -i RIGHT -filter_complex [0:v]scale=w=iw/2[left],[1:v]scale=w=iw/2[right],[left][right]framepack=sbs OUTPUT
Change the frame rate by interpolating new video output frames from the source frames.
This filter is not designed to function correctly with interlaced media. If you wish to change the frame rate of interlaced media then you are required to deinterlace before this filter and re-interlace after this filter.
A description of the accepted options follows.
Specify the output frames per second. This option can also be specified
as a value alone. The default is 50
.
Specify the start of a range where the output frame will be created as a
linear interpolation of two frames. The range is [0
-255
],
the default is 15
.
Specify the end of a range where the output frame will be created as a
linear interpolation of two frames. The range is [0
-255
],
the default is 240
.
Specify the level at which a scene change is detected as a value between
0 and 100 to indicate a new scene; a low value reflects a low
probability for the current frame to introduce a new scene, while a higher
value means the current frame is more likely to be one.
The default is 8.2
.
Specify flags influencing the filter process.
Available value for flags is:
Enable scene change detection using the value of the option scene. This flag is enabled by default.
Select one frame every N-th frame.
This filter accepts the following option:
Select frame after every step
frames.
Allowed values are positive integers higher than 0. Default value is 1
.
Detect frozen video.
This filter logs a message and sets frame metadata when it detects that the input video has no significant change in content during a specified duration. Video freeze detection calculates the mean average absolute difference of all the components of video frames and compares it to a noise floor.
The printed times and duration are expressed in seconds. The
lavfi.freezedetect.freeze_start
metadata key is set on the first frame
whose timestamp equals or exceeds the detection duration and it contains the
timestamp of the first frame of the freeze. The
lavfi.freezedetect.freeze_duration
and
lavfi.freezedetect.freeze_end
metadata keys are set on the first frame
after the freeze.
The filter accepts the following options:
Set noise tolerance. Can be specified in dB (in case "dB" is appended to the specified value) or as a difference ratio between 0 and 1. Default is -60dB, or 0.001.
Set freeze duration until notification (default is 2 seconds).
Freeze video frames.
This filter freezes video frames using frame from 2nd input.
The filter accepts the following options:
Set number of first frame from which to start freeze.
Set number of last frame from which to end freeze.
Set number of frame from 2nd input which will be used instead of replaced frames.
Apply a frei0r effect to the input video.
To enable the compilation of this filter, you need to install the frei0r
header and configure FFmpeg with --enable-frei0r
.
It accepts the following parameters:
The name of the frei0r effect to load. If the environment variable
FREI0R_PATH
is defined, the frei0r effect is searched for in each of the
directories specified by the colon-separated list in FREI0R_PATH
.
Otherwise, the standard frei0r paths are searched, in this order:
HOME/.frei0r-1/lib/, /usr/local/lib/frei0r-1/,
/usr/lib/frei0r-1/.
A ’|’-separated list of parameters to pass to the frei0r effect.
A frei0r effect parameter can be a boolean (its value is either "y" or "n"), a double, a color (specified as R/G/B, where R, G, and B are floating point numbers between 0.0 and 1.0, inclusive) or a color description as specified in the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual, a position (specified as X/Y, where X and Y are floating point numbers) and/or a string.
The number and types of parameters depend on the loaded effect. If an effect parameter is not specified, the default value is set.
frei0r=filter_name=distort0r:filter_params=0.5|0.01
frei0r=colordistance:0.2/0.3/0.4 frei0r=colordistance:violet frei0r=colordistance:0x112233
frei0r=perspective:0.2/0.2|0.8/0.2
For more information, see http://frei0r.dyne.org
This filter supports the filter_params option as commands.
Apply fast and simple postprocessing. It is a faster version of spp.
It splits (I)DCT into horizontal/vertical passes. Unlike the simple post- processing filter, one of them is performed once per block, not per pixel. This allows for much higher speed.
The filter accepts the following options:
Set quality. This option defines the number of levels for averaging. It accepts
an integer in the range 4-5. Default value is 4
.
Force a constant quantization parameter. It accepts an integer in range 0-63. If not set, the filter will use the QP from the video stream (if available).
Set filter strength. It accepts an integer in range -15 to 32. Lower values mean
more details but also more artifacts, while higher values make the image smoother
but also blurrier. Default value is 0
− PSNR optimal.
Enable the use of the QP from the B-Frames if set to 1
. Using this
option may cause flicker since the B-Frames have often larger QP. Default is
0
(not enabled).
Apply Gaussian blur filter.
The filter accepts the following options:
Set horizontal sigma, standard deviation of Gaussian blur. Default is 0.5
.
Set number of steps for Gaussian approximation. Default is 1
.
Set which planes to filter. By default all planes are filtered.
Set vertical sigma, if negative it will be same as sigma
.
Default is -1
.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Apply generic equation to each pixel.
The filter accepts the following options:
Set the luma expression.
Set the chrominance blue expression.
Set the chrominance red expression.
Set the alpha expression.
Set the red expression.
Set the green expression.
Set the blue expression.
The colorspace is selected according to the specified options. If one of the lum_expr, cb_expr, or cr_expr options is specified, the filter will automatically select a YCbCr colorspace. If one of the red_expr, green_expr, or blue_expr options is specified, it will select an RGB colorspace.
If one of the chrominance expression is not defined, it falls back on the other one. If no alpha expression is specified it will evaluate to opaque value. If none of chrominance expressions are specified, they will evaluate to the luma expression.
The expressions can use the following variables and functions:
The sequential number of the filtered frame, starting from 0
.
The coordinates of the current sample.
The width and height of the image.
Width and height scale depending on the currently filtered plane. It is the
ratio between the corresponding luma plane number of pixels and the current
plane ones. E.g. for YUV4:2:0 the values are 1,1
for the luma plane, and
0.5,0.5
for chroma planes.
Time of the current frame, expressed in seconds.
Return the value of the pixel at location (x,y) of the current plane.
Return the value of the pixel at location (x,y) of the luma plane.
Return the value of the pixel at location (x,y) of the blue-difference chroma plane. Return 0 if there is no such plane.
Return the value of the pixel at location (x,y) of the red-difference chroma plane. Return 0 if there is no such plane.
Return the value of the pixel at location (x,y) of the red/green/blue component. Return 0 if there is no such component.
Return the value of the pixel at location (x,y) of the alpha plane. Return 0 if there is no such plane.
Sum of sample values in the rectangle from (0,0) to (x,y), this allows obtaining sums of samples within a rectangle. See the functions without the sum postfix.
Set one of interpolation methods:
Default is bilinear.
For functions, if x and y are outside the area, the value will be automatically clipped to the closer edge.
Please note that this filter can use multiple threads in which case each slice will have its own expression state. If you want to use only a single expression state because your expressions depend on previous state then you should limit the number of filter threads to 1.
geq=p(W-X\,Y)
PI/3
and a
wavelength of 100 pixels:
geq=128 + 100*sin(2*(PI/100)*(cos(PI/3)*(X-50*T) + sin(PI/3)*Y)):128:128
nullsrc=s=256x256,geq=random(1)/hypot(X-cos(N*0.07)*W/2-W/2\,Y-sin(N*0.09)*H/2-H/2)^2*1000000*sin(N*0.02):128:128
format=gray,geq=lum_expr='(p(X,Y)+(256-p(X-4,Y-4)))/2'
geq=r='X/W*r(X,Y)':g='(1-X/W)*g(X,Y)':b='(H-Y)/H*b(X,Y)'
geq=lum=255*gauss((X/W-0.5)*3)*gauss((Y/H-0.5)*3)/gauss(0)/gauss(0),format=gray
Fix the banding artifacts that are sometimes introduced into nearly flat regions by truncation to 8-bit color depth. Interpolate the gradients that should go where the bands are, and dither them.
It is designed for playback only. Do not use it prior to lossy compression, because compression tends to lose the dither and bring back the bands.
It accepts the following parameters:
The maximum amount by which the filter will change any one pixel. This is also the threshold for detecting nearly flat regions. Acceptable values range from .51 to 64; the default value is 1.2. Out-of-range values will be clipped to the valid range.
The neighborhood to fit the gradient to. A larger radius makes for smoother gradients, but also prevents the filter from modifying the pixels near detailed regions. Acceptable values are 8-32; the default value is 16. Out-of-range values will be clipped to the valid range.
Alternatively, the options can be specified as a flat string: strength[:radius]
Show various filtergraph stats.
With this filter one can debug complete filtergraph. Especially issues with links filling with queued frames.
The filter accepts the following options:
Set video output size. Default is hd720.
Set video opacity. Default is 0.9. Allowed range is from 0 to 1.
Set output mode flags.
Available values for flags are:
No any filtering. Default.
Show only filters with queued frames.
Show only filters with non-zero stats.
Show only filters with non-eof stat.
Show only filters that are enabled in timeline.
Set flags which enable which stats are shown in video.
Available values for flags are:
All flags turned off.
All flags turned on.
Display number of queued frames in each link.
Display number of frames taken from filter.
Display number of frames given out from filter.
Display delta number of frames between above two values.
Display current filtered frame pts.
Display pts delta between current and previous frame.
Display current filtered frame time.
Display time delta between current and previous frame.
Display time base for filter link.
Display used format for filter link.
Display video size or number of audio channels in case of audio used by filter link.
Display video frame rate or sample rate in case of audio used by filter link.
Display link output status.
Display number of samples taken from filter.
Display number of samples given out from filter.
Display delta number of samples between above two values.
Show the timeline filter status.
Set upper limit for video rate of output stream, Default value is 25. This guarantee that output video frame rate will not be higher than this value.
A color constancy filter that applies color correction based on the grayworld assumption
The algorithm uses linear light, so input data should be linearized beforehand (and possibly correctly tagged).
ffmpeg -i INPUT -vf zscale=transfer=linear,grayworld,zscale=transfer=bt709,format=yuv420p OUTPUT
A color constancy variation filter which estimates scene illumination via grey edge algorithm and corrects the scene colors accordingly.
See: https://staff.science.uva.nl/th.gevers/pub/GeversTIP07.pdf
The filter accepts the following options:
The order of differentiation to be applied on the scene. Must be chosen in the range [0,2] and default value is 1.
The Minkowski parameter to be used for calculating the Minkowski distance. Must be chosen in the range [0,20] and default value is 1. Set to 0 for getting max value instead of calculating Minkowski distance.
The standard deviation of Gaussian blur to be applied on the scene. Must be chosen in the range [0,1024.0] and default value = 1. floor( sigma * break_off_sigma(3) ) can’t be equal to 0 if difford is greater than 0.
greyedge=difford=1:minknorm=5:sigma=2
greyedge=difford=1:minknorm=0:sigma=2
Apply guided filter for edge-preserving smoothing, dehazing and so on.
The filter accepts the following options:
Set the box radius in pixels. Allowed range is 1 to 20. Default is 3.
Set regularization parameter (with square). Allowed range is 0 to 1. Default is 0.01.
Set filter mode. Can be basic
or fast
.
Default is basic
.
Set subsampling ratio for fast
mode.
Range is 2 to 64. Default is 4.
No subsampling occurs in basic
mode.
Set guidance mode. Can be off
or on
. Default is off
.
If off
, single input is required.
If on
, two inputs of the same resolution and pixel format are required.
The second input serves as the guidance.
Set planes to filter. Default is first only.
This filter supports the all above options as commands.
ffmpeg -i in.png -vf guided out.png
ffmpeg -i in.png -i guidance.png -filter_complex guided=guidance=on out.png
Apply a Hald CLUT to a video stream.
First input is the video stream to process, and second one is the Hald CLUT. The Hald CLUT input can be a simple picture or a complete video stream.
The filter accepts the following options:
Set which CLUT video frames will be processed from second input stream, can be first or all. Default is all.
Force termination when the shortest input terminates. Default is 0
.
Continue applying the last CLUT after the end of the stream. A value of
0
disable the filter after the last frame of the CLUT is reached.
Default is 1
.
haldclut
also has the same interpolation options as lut3d (both
filters share the same internals).
This filter also supports the framesync options.
More information about the Hald CLUT can be found on Eskil Steenberg’s website (Hald CLUT author) at http://www.quelsolaar.com/technology/clut.html.
This filter supports the interp
option as commands.
Generate an identity Hald CLUT stream altered with various effects:
ffmpeg -f lavfi -i haldclutsrc=8 -vf "hue=H=2*PI*t:s=sin(2*PI*t)+1, curves=cross_process" -t 10 -c:v ffv1 clut.nut
Note: make sure you use a lossless codec.
Then use it with haldclut
to apply it on some random stream:
ffmpeg -f lavfi -i mandelbrot -i clut.nut -filter_complex '[0][1] haldclut' -t 20 mandelclut.mkv
The Hald CLUT will be applied to the 10 first seconds (duration of
clut.nut), then the latest picture of that CLUT stream will be applied
to the remaining frames of the mandelbrot
stream.
A Hald CLUT is supposed to be a squared image of Level*Level*Level
by
Level*Level*Level
pixels. For a given Hald CLUT, FFmpeg will select the
biggest possible square starting at the top left of the picture. The remaining
padding pixels (bottom or right) will be ignored. This area can be used to add
a preview of the Hald CLUT.
Typically, the following generated Hald CLUT will be supported by the
haldclut
filter:
ffmpeg -f lavfi -i haldclutsrc=8 -vf " pad=iw+320 [padded_clut]; smptebars=s=320x256, split [a][b]; [padded_clut][a] overlay=W-320:h, curves=color_negative [main]; [main][b] overlay=W-320" -frames:v 1 clut.png
It contains the original and a preview of the effect of the CLUT: SMPTE color bars are displayed on the right-top, and below the same color bars processed by the color changes.
Then, the effect of this Hald CLUT can be visualized with:
ffplay input.mkv -vf "movie=clut.png, [in] haldclut"
Flip the input video horizontally.
For example, to horizontally flip the input video with ffmpeg
:
ffmpeg -i in.avi -vf "hflip" out.avi
This filter applies a global color histogram equalization on a per-frame basis.
It can be used to correct video that has a compressed range of pixel intensities. The filter redistributes the pixel intensities to equalize their distribution across the intensity range. It may be viewed as an "automatically adjusting contrast filter". This filter is useful only for correcting degraded or poorly captured source video.
The filter accepts the following options:
Determine the amount of equalization to be applied. As the strength is reduced, the distribution of pixel intensities more-and-more approaches that of the input frame. The value must be a float number in the range [0,1] and defaults to 0.200.
Set the maximum intensity that can generated and scale the output values appropriately. The strength should be set as desired and then the intensity can be limited if needed to avoid washing-out. The value must be a float number in the range [0,1] and defaults to 0.210.
Set the antibanding level. If enabled the filter will randomly vary
the luminance of output pixels by a small amount to avoid banding of
the histogram. Possible values are none
, weak
or
strong
. It defaults to none
.
Compute and draw a color distribution histogram for the input video.
The computed histogram is a representation of the color component distribution in an image.
Standard histogram displays the color components distribution in an image. Displays color graph for each color component. Shows distribution of the Y, U, V, A or R, G, B components, depending on input format, in the current frame. Below each graph a color component scale meter is shown.
The filter accepts the following options:
Set height of level. Default value is 200
.
Allowed range is [50, 2048].
Set height of color scale. Default value is 12
.
Allowed range is [0, 40].
Set display mode. It accepts the following values:
Per color component graphs are placed below each other.
Per color component graphs are placed side by side.
Presents information identical to that in the parade
, except
that the graphs representing color components are superimposed directly
over one another.
Default is stack
.
Set mode. Can be either linear
, or logarithmic
.
Default is linear
.
Set what color components to display.
Default is 7
.
Set foreground opacity. Default is 0.7
.
Set background opacity. Default is 0.5
.
Set colors mode. It accepts the following values:
Default is whiteonblack
.
This is a high precision/quality 3d denoise filter. It aims to reduce image noise, producing smooth images and making still images really still. It should enhance compressibility.
It accepts the following optional parameters:
A non-negative floating point number which specifies spatial luma strength. It defaults to 4.0.
A non-negative floating point number which specifies spatial chroma strength. It defaults to 3.0*luma_spatial/4.0.
A floating point number which specifies luma temporal strength. It defaults to 6.0*luma_spatial/4.0.
A floating point number which specifies chroma temporal strength. It defaults to luma_tmp*chroma_spatial/luma_spatial.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Download hardware frames to system memory.
The input must be in hardware frames, and the output a non-hardware format. Not all formats will be supported on the output - it may be necessary to insert an additional format filter immediately following in the graph to get the output in a supported format.
Map hardware frames to system memory or to another device.
This filter has several different modes of operation; which one is used depends on the input and output formats:
Map the input frames to system memory and pass them to the output. If the original hardware frame is later required (for example, after overlaying something else on part of it), the hwmap filter can be used again in the next mode to retrieve it.
If the input is actually a software-mapped hardware frame, then unmap it - that is, return the original hardware frame.
Otherwise, a device must be provided. Create new hardware surfaces on that device for the output, then map them back to the software format at the input and give those frames to the preceding filter. This will then act like the hwupload filter, but may be able to avoid an additional copy when the input is already in a compatible format.
A device must be supplied for the output, either directly or with the derive_device option. The input and output devices must be of different types and compatible - the exact meaning of this is system-dependent, but typically it means that they must refer to the same underlying hardware context (for example, refer to the same graphics card).
If the input frames were originally created on the output device, then unmap to retrieve the original frames.
Otherwise, map the frames to the output device - create new hardware frames on the output corresponding to the frames on the input.
The following additional parameters are accepted:
Set the frame mapping mode. Some combination of:
The mapped frame should be readable.
The mapped frame should be writeable.
The mapping will always overwrite the entire frame.
This may improve performance in some cases, as the original contents of the frame need not be loaded.
The mapping must not involve any copying.
Indirect mappings to copies of frames are created in some cases where either direct mapping is not possible or it would have unexpected properties. Setting this flag ensures that the mapping is direct and will fail if that is not possible.
Defaults to read+write if not specified.
Rather than using the device supplied at initialisation, instead derive a new device of type type from the device the input frames exist on.
In a hardware to hardware mapping, map in reverse - create frames in the sink and map them back to the source. This may be necessary in some cases where a mapping in one direction is required but only the opposite direction is supported by the devices being used.
This option is dangerous - it may break the preceding filter in undefined ways if there are any additional constraints on that filter’s output. Do not use it without fully understanding the implications of its use.
Upload system memory frames to hardware surfaces.
The device to upload to must be supplied when the filter is initialised. If using ffmpeg, select the appropriate device with the -filter_hw_device option or with the derive_device option. The input and output devices must be of different types and compatible - the exact meaning of this is system-dependent, but typically it means that they must refer to the same underlying hardware context (for example, refer to the same graphics card).
The following additional parameters are accepted:
Rather than using the device supplied at initialisation, instead derive a new device of type type from the device the input frames exist on.
Upload system memory frames to a CUDA device.
It accepts the following optional parameters:
The number of the CUDA device to use
Apply a high-quality magnification filter designed for pixel art. This filter was originally created by Maxim Stepin.
It accepts the following option:
Set the scaling dimension: 2
for hq2x
, 3
for
hq3x
and 4
for hq4x
.
Default is 3
.
Stack input videos horizontally.
All streams must be of same pixel format and of same height.
Note that this filter is faster than using overlay and pad filter to create same output.
The filter accepts the following option:
Set number of input streams. Default is 2.
If set to 1, force the output to terminate when the shortest input terminates. Default value is 0.
Turns a certain HSV range into gray values.
This filter measures color difference between set HSV color in options and ones measured in video stream. Depending on options, output colors can be changed to be gray or not.
The filter accepts the following options:
Set the hue value which will be used in color difference calculation. Allowed range is from -360 to 360. Default value is 0.
Set the saturation value which will be used in color difference calculation. Allowed range is from -1 to 1. Default value is 0.
Set the value which will be used in color difference calculation. Allowed range is from -1 to 1. Default value is 0.
Set similarity percentage with the key color. Allowed range is from 0 to 1. Default value is 0.01.
0.00001 matches only the exact key color, while 1.0 matches everything.
Blend percentage. Allowed range is from 0 to 1. Default value is 0.
0.0 makes pixels either fully gray, or not gray at all.
Higher values result in more gray pixels, with a higher gray pixel the more similar the pixels color is to the key color.
Turns a certain HSV range into transparency.
This filter measures color difference between set HSV color in options and ones measured in video stream. Depending on options, output colors can be changed to transparent by adding alpha channel.
The filter accepts the following options:
Set the hue value which will be used in color difference calculation. Allowed range is from -360 to 360. Default value is 0.
Set the saturation value which will be used in color difference calculation. Allowed range is from -1 to 1. Default value is 0.
Set the value which will be used in color difference calculation. Allowed range is from -1 to 1. Default value is 0.
Set similarity percentage with the key color. Allowed range is from 0 to 1. Default value is 0.01.
0.00001 matches only the exact key color, while 1.0 matches everything.
Blend percentage. Allowed range is from 0 to 1. Default value is 0.
0.0 makes pixels either fully transparent, or not transparent at all.
Higher values result in semi-transparent pixels, with a higher transparency the more similar the pixels color is to the key color.
Modify the hue and/or the saturation of the input.
It accepts the following parameters:
Specify the hue angle as a number of degrees. It accepts an expression, and defaults to "0".
Specify the saturation in the [-10,10] range. It accepts an expression and defaults to "1".
Specify the hue angle as a number of radians. It accepts an expression, and defaults to "0".
Specify the brightness in the [-10,10] range. It accepts an expression and defaults to "0".
h and H are mutually exclusive, and can’t be specified at the same time.
The b, h, H and s option values are expressions containing the following constants:
frame count of the input frame starting from 0
presentation timestamp of the input frame expressed in time base units
frame rate of the input video, NAN if the input frame rate is unknown
timestamp expressed in seconds, NAN if the input timestamp is unknown
time base of the input video
hue=h=90:s=1
hue=H=PI/2:s=1
hue="H=2*PI*t: s=sin(2*PI*t)+1"
hue="s=min(t/3\,1)"
The general fade-in expression can be written as:
hue="s=min(0\, max((t-START)/DURATION\, 1))"
hue="s=max(0\, min(1\, (8-t)/3))"
The general fade-out expression can be written as:
hue="s=max(0\, min(1\, (START+DURATION-t)/DURATION))"
This filter supports the following commands:
Modify the hue and/or the saturation and/or brightness of the input video. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Apply hue-saturation-intensity adjustments to input video stream.
This filter operates in RGB colorspace.
This filter accepts the following options:
Set the hue shift in degrees to apply. Default is 0. Allowed range is from -180 to 180.
Set the saturation shift. Default is 0. Allowed range is from -1 to 1.
Set the intensity shift. Default is 0. Allowed range is from -1 to 1.
Set which primary and complementary colors are going to be adjusted. This options is set by providing one or multiple values. This can select multiple colors at once. By default all colors are selected.
Adjust reds.
Adjust yellows.
Adjust greens.
Adjust cyans.
Adjust blues.
Adjust magentas.
Adjust all colors.
Set strength of filtering. Allowed range is from 0 to 100. Default value is 1.
Set weight for each RGB component. Allowed range is from 0 to 1. By default is set to 0.333, 0.334, 0.333. Those options are used in saturation and lightess processing.
Set preserving lightness, by default is disabled. Adjusting hues can change lightness from original RGB triplet, with this option enabled lightness is kept at same value.
Grow first stream into second stream by connecting components. This makes it possible to build more robust edge masks.
This filter accepts the following options:
Set which planes will be processed as bitmap, unprocessed planes will be copied from first stream. By default value 0xf, all planes will be processed.
Set threshold which is used in filtering. If pixel component value is higher than this value filter algorithm for connecting components is activated. By default value is 0.
The hysteresis
filter also supports the framesync options.
Detect the colorspace from an embedded ICC profile (if present), and update the frame’s tags accordingly.
This filter accepts the following options:
If true, the frame’s existing colorspace tags will always be overridden by
values detected from an ICC profile. Otherwise, they will only be assigned if
they contain unknown
. Enabled by default.
Generate ICC profiles and attach them to frames.
This filter accepts the following options:
Configure the colorspace that the ICC profile will be generated for. The
default value of auto
infers the value from the input frame’s metadata,
defaulting to BT.709/sRGB as appropriate.
See the setparams filter for a list of possible values, but note that
unknown
are not valid values for this filter.
If true, an ICC profile will be generated even if it would overwrite an already existing ICC profile. Disabled by default.
Obtain the identity score between two input videos.
This filter takes two input videos.
Both input videos must have the same resolution and pixel format for this filter to work correctly. Also it assumes that both inputs have the same number of frames, which are compared one by one.
The obtained per component, average, min and max identity score is printed through the logging system.
The filter stores the calculated identity scores of each frame in frame metadata.
This filter also supports the framesync options.
In the below example the input file main.mpg being processed is compared with the reference file ref.mpg.
ffmpeg -i main.mpg -i ref.mpg -lavfi identity -f null -
Detect video interlacing type.
This filter tries to detect if the input frames are interlaced, progressive, top or bottom field first. It will also try to detect fields that are repeated between adjacent frames (a sign of telecine).
Single frame detection considers only immediately adjacent frames when classifying each frame. Multiple frame detection incorporates the classification history of previous frames.
The filter will log these metadata values:
Detected type of current frame using single-frame detection. One of: “tff” (top field first), “bff” (bottom field first), “progressive”, or “undetermined”
Cumulative number of frames detected as top field first using single-frame detection.
Cumulative number of frames detected as top field first using multiple-frame detection.
Cumulative number of frames detected as bottom field first using single-frame detection.
Detected type of current frame using multiple-frame detection. One of: “tff” (top field first), “bff” (bottom field first), “progressive”, or “undetermined”
Cumulative number of frames detected as bottom field first using multiple-frame detection.
Cumulative number of frames detected as progressive using single-frame detection.
Cumulative number of frames detected as progressive using multiple-frame detection.
Cumulative number of frames that could not be classified using single-frame detection.
Cumulative number of frames that could not be classified using multiple-frame detection.
Which field in the current frame is repeated from the last. One of “neither”, “top”, or “bottom”.
Cumulative number of frames with no repeated field.
Cumulative number of frames with the top field repeated from the previous frame’s top field.
Cumulative number of frames with the bottom field repeated from the previous frame’s bottom field.
The filter accepts the following options:
Set interlacing threshold.
Set progressive threshold.
Threshold for repeated field detection.
Number of frames after which a given frame’s contribution to the statistics is halved (i.e., it contributes only 0.5 to its classification). The default of 0 means that all frames seen are given full weight of 1.0 forever.
When this is not 0 then idet will use the specified number of frames to determine if the interlaced flag is accurate, it will not count undetermined frames. If the flag is found to be accurate it will be used without any further computations, if it is found to be inaccurate it will be cleared without any further computations. This allows inserting the idet filter as a low computational method to clean up the interlaced flag
Deinterleave or interleave fields.
This filter allows one to process interlaced images fields without deinterlacing them. Deinterleaving splits the input frame into 2 fields (so called half pictures). Odd lines are moved to the top half of the output image, even lines to the bottom half. You can process (filter) them independently and then re-interleave them.
The filter accepts the following options:
Available values for luma_mode, chroma_mode and alpha_mode are:
Do nothing.
Deinterleave fields, placing one above the other.
Interleave fields. Reverse the effect of deinterleaving.
Default value is none
.
Swap luma/chroma/alpha fields. Exchange even & odd lines. Default value is 0
.
This filter supports the all above options as commands.
Apply inflate effect to the video.
This filter replaces the pixel by the local(3x3) average by taking into account only values higher than the pixel.
It accepts the following options:
Limit the maximum change for each plane, default is 65535. If 0, plane will remain unchanged.
This filter supports the all above options as commands.
Simple interlacing filter from progressive contents. This interleaves upper (or lower) lines from odd frames with lower (or upper) lines from even frames, halving the frame rate and preserving image height.
Original Original New Frame Frame 'j' Frame 'j+1' (tff) ========== =========== ================== Line 0 --------------------> Frame 'j' Line 0 Line 1 Line 1 ----> Frame 'j+1' Line 1 Line 2 ---------------------> Frame 'j' Line 2 Line 3 Line 3 ----> Frame 'j+1' Line 3 ... ... ... New Frame + 1 will be generated by Frame 'j+2' and Frame 'j+3' and so on
It accepts the following optional parameters:
This determines whether the interlaced frame is taken from the even (tff - default) or odd (bff) lines of the progressive frame.
Vertical lowpass filter to avoid twitter interlacing and reduce moire patterns.
Disable vertical lowpass filter
Enable linear filter (default)
Enable complex filter. This will slightly less reduce twitter and moire but better retain detail and subjective sharpness impression.
Deinterlace input video by applying Donald Graft’s adaptive kernel deinterling. Work on interlaced parts of a video to produce progressive frames.
The description of the accepted parameters follows.
Set the threshold which affects the filter’s tolerance when determining if a pixel line must be processed. It must be an integer in the range [0,255] and defaults to 10. A value of 0 will result in applying the process on every pixels.
Paint pixels exceeding the threshold value to white if set to 1. Default is 0.
Set the fields order. Swap fields if set to 1, leave fields alone if 0. Default is 0.
Enable additional sharpening if set to 1. Default is 0.
Enable twoway sharpening if set to 1. Default is 0.
kerndeint=thresh=10:map=0:order=0:sharp=0:twoway=0
kerndeint=sharp=1
kerndeint=map=1
Apply kirsch operator to input video stream.
The filter accepts the following option:
Set which planes will be processed, unprocessed planes will be copied. By default value 0xf, all planes will be processed.
Set value which will be multiplied with filtered result.
Set value which will be added to filtered result.
This filter supports the all above options as commands.
Slowly update darker pixels.
This filter makes short flashes of light appear longer. This filter accepts the following options:
Set factor for decaying. Default is .95. Allowed range is from 0 to 1.
Set which planes to filter. Default is all. Allowed range is from 0 to 15.
This filter supports the all above options as commands.
Correct radial lens distortion
This filter can be used to correct for radial distortion as can result from the use of wide angle lenses, and thereby re-rectify the image. To find the right parameters one can use tools available for example as part of opencv or simply trial-and-error. To use opencv use the calibration sample (under samples/cpp) from the opencv sources and extract the k1 and k2 coefficients from the resulting matrix.
Note that effectively the same filter is available in the open-source tools Krita and Digikam from the KDE project.
In contrast to the vignette filter, which can also be used to compensate lens errors, this filter corrects the distortion of the image, whereas vignette corrects the brightness distribution, so you may want to use both filters together in certain cases, though you will have to take care of ordering, i.e. whether vignetting should be applied before or after lens correction.
The filter accepts the following options:
Relative x-coordinate of the focal point of the image, and thereby the center of the distortion. This value has a range [0,1] and is expressed as fractions of the image width. Default is 0.5.
Relative y-coordinate of the focal point of the image, and thereby the center of the distortion. This value has a range [0,1] and is expressed as fractions of the image height. Default is 0.5.
Coefficient of the quadratic correction term. This value has a range [-1,1]. 0 means no correction. Default is 0.
Coefficient of the double quadratic correction term. This value has a range [-1,1]. 0 means no correction. Default is 0.
Set interpolation type. Can be nearest
or bilinear
.
Default is nearest
.
Specify the color of the unmapped pixels. For the syntax of this option,
check the (ffmpeg-utils)"Color" section in the ffmpeg-utils
manual. Default color is black@0
.
The formula that generates the correction is:
r_src = r_tgt * (1 + k1 * (r_tgt / r_0)^2 + k2 * (r_tgt / r_0)^4)
where r_0 is halve of the image diagonal and r_src and r_tgt are the distances from the focal point in the source and target images, respectively.
This filter supports the all above options as commands.
Apply lens correction via the lensfun library (http://lensfun.sourceforge.net/).
The lensfun
filter requires the camera make, camera model, and lens model
to apply the lens correction. The filter will load the lensfun database and
query it to find the corresponding camera and lens entries in the database. As
long as these entries can be found with the given options, the filter can
perform corrections on frames. Note that incomplete strings will result in the
filter choosing the best match with the given options, and the filter will
output the chosen camera and lens models (logged with level "info"). You must
provide the make, camera model, and lens model as they are required.
To obtain a list of available makes and models, leave out one or both of make
and
model
options. The filter will send the full list to the log with level INFO
.
The first column is the make and the second column is the model.
To obtain a list of available lenses, set any values for make and model and leave out the
lens_model
option. The filter will send the full list of lenses in the log with level
INFO
. The ffmpeg tool will exit after the list is printed.
The filter accepts the following options:
The make of the camera (for example, "Canon"). This option is required.
The model of the camera (for example, "Canon EOS 100D"). This option is required.
The model of the lens (for example, "Canon EF-S 18-55mm f/3.5-5.6 IS STM"). This option is required.
The full path to the lens database folder. If not set, the filter will attempt to load the database from the install path when the library was built. Default is unset.
The type of correction to apply. The following values are valid options:
Enables fixing lens vignetting.
Enables fixing lens geometry. This is the default.
Enables fixing chromatic aberrations.
Enables fixing lens vignetting and lens geometry.
Enables fixing lens vignetting and chromatic aberrations.
Enables fixing both lens geometry and chromatic aberrations.
Enables all possible corrections.
The focal length of the image/video (zoom; expected constant for video). For example, a 18–55mm lens has focal length range of [18–55], so a value in that range should be chosen when using that lens. Default 18.
The aperture of the image/video (expected constant for video). Note that aperture is only used for vignetting correction. Default 3.5.
The focus distance of the image/video (expected constant for video). Note that focus distance is only used for vignetting and only slightly affects the vignetting correction process. If unknown, leave it at the default value (which is 1000).
The scale factor which is applied after transformation. After correction the video is no longer necessarily rectangular. This parameter controls how much of the resulting image is visible. The value 0 means that a value will be chosen automatically such that there is little or no unmapped area in the output image. 1.0 means that no additional scaling is done. Lower values may result in more of the corrected image being visible, while higher values may avoid unmapped areas in the output.
The target geometry of the output image/video. The following values are valid options:
Apply the reverse of image correction (instead of correcting distortion, apply it).
The type of interpolation used when correcting distortion. The following values are valid options:
ffmpeg -i input.mov -vf lensfun=make=Canon:model="Canon EOS 100D":lens_model="Canon EF-S 18-55mm f/3.5-5.6 IS STM":focal_length=18:aperture=8 -c:v h264 -b:v 8000k output.mov
ffmpeg -i input.mov -vf lensfun=make=Canon:model="Canon EOS 100D":lens_model="Canon EF-S 18-55mm f/3.5-5.6 IS STM":focal_length=18:aperture=8:enable='lte(t\,5)' -c:v h264 -b:v 8000k output.mov
Flexible GPU-accelerated processing filter based on libplacebo (https://code.videolan.org/videolan/libplacebo).
The options for this filter are divided into the following sections:
These options control the overall output mode. By default, libplacebo will try to preserve the source colorimetry and size as best as it can, but it will apply any embedded film grain, dolby vision metadata or anamorphic SAR present in source frames.
Set the number of inputs. This can be used, alongside the idx
variable,
to allow placing/blending multiple inputs inside the output frame. This
effectively enables functionality similar to hstack, overlay, etc.
Set the output video dimension expression. Default values are iw
and
ih
.
Allows for the same expressions as the scale filter.
Set the input crop x/y expressions, default values are (iw-cw)/2
and
(ih-ch)/2
.
Set the input crop width/height expressions, default values are iw
and
ih
.
Set the output placement x/y expressions, default values are (ow-pw)/2
and (oh-ph)/2
.
Set the output placement width/height expressions, default values are ow
and oh
.
Set the output frame rate. This can be rational, e.g. 60000/1001
. If
set to the special string none
(the default), input timestamps will
instead be passed through to the output unmodified. Otherwise, the input video
frames will be interpolated as necessary to rescale the video to the specified
target framerate, in a manner as determined by the frame_mixer option.
Set the output format override. If unset (the default), frames will be output in the same format as the respective input frames. Otherwise, format conversion will be performed.
Work the same as the identical scale filter options.
If enabled, output frames will always have a pixel aspect ratio of 1:1. This will introduce additional padding/cropping as necessary. If disabled (the default), any aspect ratio mismatches, including those from e.g. anamorphic video sources, are forwarded to the output pixel aspect ratio.
Specifies a ratio (between 0.0
and 1.0
) between padding and
cropping when the input aspect ratio does not match the output aspect ratio and
normalize_sar is in effect. The default of 0.0
always pads the
content with black borders, while a value of 1.0
always crops off parts
of the content. Intermediate values are possible, leading to a mix of the two
approaches.
Set the color used to fill the output area not covered by the output image, for
example as a result of normalize_sar. For the general syntax of this
option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils
manual. Defaults to black
.
Render frames with rounded corners. The value, given as a float ranging from
0.0
to 1.0
, indicates the relative degree of rounding, from fully
square to fully circular. In other words, it gives the radius divided by half
the smaller side length. Defaults to 0.0
.
Pass extra libplacebo internal configuration options. These can be specified as a list of key=value pairs separated by ’:’. The following example shows how to configure a custom filter kernel ("EWA LanczosSharp") and use it to double the input image resolution:
-vf "libplacebo=w=iw*2:h=ih*2:extra_opts='upscaler=custom\:upscaler_preset=ewa_lanczos\:upscaler_blur=0.9812505644269356'"
Configure the colorspace that output frames will be delivered in. The default
value of auto
outputs frames in the same format as the input frames,
leading to no change. For any other value, conversion will be performed.
See the setparams filter for a list of possible values.
Apply film grain (e.g. AV1 or H.274) if present in source frames, and strip it from the output. Enabled by default.
Apply Dolby Vision RPU metadata if present in source frames, and strip it from
the output. Enabled by default. Note that Dolby Vision will always output
BT.2020+PQ, overriding the usual input frame metadata. These will also be
picked as the values of auto
for the respective frame output options.
In addition to the expression constants documented for the scale filter, the crop_w, crop_h, crop_x, crop_y, pos_w, pos_h, pos_x and pos_y options can also contain the following constants:
The (0-based) numeric index of the currently active input stream.
The computed values of crop_w and crop_h.
The computed values of pos_w and pos_h.
The input frame timestamp, in seconds. NAN if input timestamp is unknown.
The input frame timestamp, in seconds. NAN if input timestamp is unknown.
The input frame number, starting with 0.
The options in this section control how libplacebo performs upscaling and (if
necessary) downscaling. Note that libplacebo will always internally operate on
4:4:4 content, so any sub-sampled chroma formats such as yuv420p
will
necessarily be upsampled and downsampled as part of the rendering process. That
means scaling might be in effect even if the source and destination resolution
are the same.
Configure the filter kernel used for upscaling and downscaling. The respective
defaults are spline36
and mitchell
. For a full list of possible
values, pass help
to these options. The most important values are:
Forces the use of built-in GPU texture sampling (typically bilinear). Extremely fast but poor quality, especially when downscaling.
Bilinear interpolation. Can generally be done for free on GPUs, except when doing so would lead to aliasing. Fast and low quality.
Nearest-neighbour interpolation. Sharp but highly aliasing.
Algorithm that looks visually similar to nearest-neighbour interpolation but tries to preserve pixel aspect ratio. Good for pixel art, since it results in minimal distortion of the artistic appearance.
Standard sinc-sinc interpolation kernel.
Cubic spline approximation of lanczos. No difference in performance, but has very slightly less ringing.
Elliptically weighted average version of lanczos, based on a jinc-sinc kernel. This is also popularly referred to as just "Jinc scaling". Slow but very high quality.
Gaussian kernel. Has certain ideal mathematical properties, but subjectively very blurry.
Cubic BC spline with parameters recommended by Mitchell and Netravali. Very little ringing.
Controls the kernel used for mixing frames temporally. The default value is
none
, which disables frame mixing. For a full list of possible values,
pass help
to this option. The most important values are:
Disables frame mixing, giving a result equivalent to "nearest neighbour" semantics.
Oversamples the input video to create a "Smooth Motion"-type effect: if an output frame would exactly fall on the transition between two video frames, it is blended according to the relative overlap. This is the recommended option whenever preserving the original subjective appearance is desired.
Larger filter kernel that smoothly interpolates multiple frames in a manner designed to eliminate ringing and other artefacts as much as possible. This is the recommended option wherever maximum visual smoothness is desired.
Linear blend/fade between frames. Especially useful for constructing e.g. slideshows.
Configures the size of scaler LUTs, ranging from 1
to 256
. The
default of 0
will pick libplacebo’s internal default, typically
64
.
Enables anti-ringing (for non-EWA filters). The value (between 0.0
and
1.0
) configures the strength of the anti-ringing algorithm. May increase
aliasing if set too high. Disabled by default.
Enable sigmoidal compression during upscaling. Reduces ringing slightly. Enabled by default.
Libplacebo comes with a built-in debanding filter that is good at counteracting many common sources of banding and blocking. Turning this on is highly recommended whenever quality is desired.
Enable (fast) debanding algorithm. Disabled by default.
Number of deband iterations of the debanding algorithm. Each iteration is
performed with progressively increased radius (and diminished threshold).
Recommended values are in the range 1
to 4
. Defaults to 1
.
Debanding filter strength. Higher numbers lead to more aggressive debanding.
Defaults to 4.0
.
Debanding filter radius. A higher radius is better for slow gradients, while
a lower radius is better for steep gradients. Defaults to 16.0
.
Amount of extra output grain to add. Helps hide imperfections. Defaults to
6.0
.
A collection of subjective color controls. Not very rigorous, so the exact effect will vary somewhat depending on the input primaries and colorspace.
Brightness boost, between -1.0
and 1.0
. Defaults to 0.0
.
Contrast gain, between 0.0
and 16.0
. Defaults to 1.0
.
Saturation gain, between 0.0
and 16.0
. Defaults to 1.0
.
Hue shift in radians, between -3.14
and 3.14
. Defaults to
0.0
. This will rotate the UV subvector, defaulting to BT.709
coefficients for RGB inputs.
Gamma adjustment, between 0.0
and 16.0
. Defaults to 1.0
.
Cone model to use for color blindness simulation. Accepts any combination of
l
, m
and s
. Here are some examples:
Deuteranomaly / deuteranopia (affecting 3%-4% of the population)
Protanomaly / protanopia (affecting 1%-2% of the population)
Monochromacy (very rare)
Achromatopsy (complete loss of daytime vision, extremely rare)
Gain factor for the cones specified by cones
, between 0.0
and
10.0
. A value of 1.0
results in no change to color vision. A
value of 0.0
(the default) simulates complete loss of those cones. Values
above 1.0
result in exaggerating the differences between cones, which
may help compensate for reduced color vision.
To help deal with sources that only have static HDR10 metadata (or no tagging whatsoever), libplacebo uses its own internal frame analysis compute shader to analyze source frames and adapt the tone mapping function in realtime. If this is too slow, or if exactly reproducible frame-perfect results are needed, it’s recommended to turn this feature off.
Enable HDR peak detection. Ignores static MaxCLL/MaxFALL values in favor of dynamic detection from the input. Note that the detected values do not get written back to the output frames, they merely guide the internal tone mapping process. Enabled by default.
Peak detection smoothing period, between 0.0
and 1000.0
. Higher
values result in peak detection becoming less responsive to changes in the
input. Defaults to 100.0
.
Lower bound on the detected peak (relative to SDR white), between 0.0
and 100.0
. Defaults to 1.0
.
Lower and upper thresholds for scene change detection. Expressed in a
logarithmic scale between 0.0
and 100.0
. Default to 5.5
and 10.0
, respectively. Setting either to a negative value disables
this functionality.
Which percentile of the frame brightness histogram to use as the source peak
for tone-mapping. Defaults to 99.995
, a fairly conservative value.
Setting this to 100.0
disables frame histogram measurement and instead
uses the true peak brightness for tone-mapping.
The options in this section control how libplacebo performs tone-mapping and gamut-mapping when dealing with mismatches between wide-gamut or HDR content. In general, libplacebo relies on accurate source tagging and mastering display gamut information to produce the best results.
How to handle out-of-gamut colors that can occur as a result of colorimetric gamut mapping.
Do nothing, simply clip out-of-range colors to the RGB volume. Low quality but extremely fast.
Perceptually soft-clip colors to the gamut volume. This is the default.
Relative colorimetric hard-clip. Similar to perceptual
but without
the soft knee.
Saturation mapping, maps primaries directly to primaries in RGB space. Not recommended except for artificial computer graphics for which a bright, saturated display is desired.
Absolute colorimetric hard-clip. Performs no adjustment of the white point.
Hard-desaturates out-of-gamut colors towards white, while preserving the luminance. Has a tendency to distort the visual appearance of bright objects.
Linearly reduces content brightness to preserves saturated details, followed by clipping the remaining out-of-gamut colors.
Highlight out-of-gamut pixels (by inverting/marking them).
Linearly reduces chromaticity of the entire image to make it fit within the target color volume. Be careful when using this on BT.2020 sources without proper mastering metadata, as doing so will lead to excessive desaturation.
Tone-mapping algorithm to use. Available values are:
Automatic selection based on internal heuristics. This is the default.
Performs no tone-mapping, just clips out-of-range colors. Retains perfect color accuracy for in-range colors but completely destroys out-of-range information. Does not perform any black point adaptation. Not configurable.
EETF from SMPTE ST 2094-40 Annex B, which applies the Bezier curves from HDR10+ dynamic metadata based on Bezier curves to perform tone-mapping. The OOTF used is adjusted based on the ratio between the targeted and actual display peak luminances.
EETF from SMPTE ST 2094-10 Annex B.2, which takes into account the input signal
average luminance in addition to the maximum/minimum. The configurable contrast
parameter influences the slope of the linear output segment, defaulting to
1.0
for no increase/decrease in contrast. Note that this does not
currently include the subjective gain/offset/gamma controls defined in Annex
B.3.
EETF from the ITU-R Report BT.2390, a hermite spline roll-off with linear
segment. The knee point offset is configurable. Note that this parameter
defaults to 1.0
, rather than the value of 0.5
from the ITU-R
spec.
EETF from ITU-R Report BT.2446, method A. Designed for well-mastered HDR sources. Can be used for both forward and inverse tone mapping. Not configurable.
Simple spline consisting of two polynomials, joined by a single pivot point.
The parameter gives the pivot point (in PQ space), defaulting to 0.30
.
Can be used for both forward and inverse tone mapping.
Simple non-linear, global tone mapping algorithm. The parameter specifies the
local contrast coefficient at the display peak. Essentially, a parameter of
0.5
implies that the reference white will be about half as bright as
when clipping. Defaults to 0.5
, which results in the simplest
formulation of this function.
Generalization of the reinhard tone mapping algorithm to support an additional
linear slope near black. The tone mapping parameter indicates the trade-off
between the linear section and the non-linear section. Essentially, for a given
parameter x, every color value below x will be mapped linearly,
while higher values get non-linearly tone-mapped. Values near 1.0
make
this curve behave like clip
, while values near 0.0
make this
curve behave like reinhard
. The default value is 0.3
, which
provides a good balance between colorimetric accuracy and preserving
out-of-gamut details.
Piece-wise, filmic tone-mapping algorithm developed by John Hable for use in
Uncharted 2, inspired by a similar tone-mapping algorithm used by Kodak.
Popularized by its use in video games with HDR rendering. Preserves both dark
and bright details very well, but comes with the drawback of changing the
average brightness quite significantly. This is sort of similar to
reinhard
with parameter 0.24
.
Fits a gamma (power) function to transfer between the source and target color
spaces, effectively resulting in a perceptual hard-knee joining two roughly
linear sections. This preserves details at all scales fairly accurately, but
can result in an image with a muted or dull appearance. The parameter is used
as the cutoff point, defaulting to 0.5
.
Linearly stretches the input range to the output range, in PQ space. This will
preserve all details accurately, but results in a significantly different
average brightness. Can be used for inverse tone-mapping in addition to regular
tone-mapping. The parameter can be used as an additional linear gain
coefficient (defaulting to 1.0
).
For tunable tone mapping functions, this parameter can be used to fine-tune the
curve behavior. Refer to the documentation of tonemapping
. The default
value of 0.0
is replaced by the curve’s preferred default setting.
If enabled, this filter will also attempt stretching SDR signals to fill HDR output color volumes. Disabled by default.
Size of the tone-mapping LUT, between 2
and 1024
. Defaults to
256
. Note that this figure is squared when combined with
peak_detect
.
Contrast recovery strength. If set to a value above 0.0
, the source
image will be divided into high-frequency and low-frequency components, and a
portion of the high-frequency image is added back onto the tone-mapped output.
May cause excessive ringing artifacts for some HDR sources, but can improve the
subjective sharpness and detail left over in the image after tone-mapping.
Defaults to 0.30
.
Contrast recovery lowpass kernel size. Defaults to 3.5
. Increasing or
decreasing this will affect the visual appearance substantially. Has no effect
when contrast_recovery
is disabled.
By default, libplacebo will dither whenever necessary, which includes rendering
to any integer format below 16-bit precision. It’s recommended to always leave
this on, since not doing so may result in visible banding in the output, even
if the debanding
filter is enabled. If maximum performance is needed,
use ordered_fixed
instead of disabling dithering.
Dithering method to use. Accepts the following values:
Disables dithering completely. May result in visible banding.
Dither with pseudo-blue noise. This is the default.
Tunable ordered dither pattern.
Faster ordered dither with a fixed size of 6
. Texture-less.
Dither with white noise. Texture-less.
Dither LUT size, as log base2 between 1
and 8
. Defaults to
6
, corresponding to a LUT size of 64x64
.
Enables temporal dithering. Disabled by default.
libplacebo supports a number of custom shaders based on the mpv .hook GLSL syntax. A collection of such shaders can be found here: https://github.com/mpv-player/mpv/wiki/User-Scripts#user-shaders
A full description of the mpv shader format is beyond the scope of this section, but a summary can be found here: https://mpv.io/manual/master/#options-glsl-shader
Specifies a path to a custom shader file to load at runtime.
Specifies a complete custom shader as a raw string.
All of the options in this section default off. They may be of assistance when attempting to squeeze the maximum performance at the cost of quality.
Disable anti-aliasing when downscaling.
Truncate polar (EWA) scaler kernels below this absolute magnitude, between
0.0
and 1.0
.
Disable linear light scaling.
Disable built-in GPU sampling (forces LUT).
Forcibly disable FBOs, resulting in loss of almost all functionality, but offering the maximum possible speed.
This filter supports almost all of the above options as commands.
libplacebo=colorspace=bt709:color_primaries=bt709:color_trc=bt709:range=tv
libplacebo=w=1920:h=1080:force_original_aspect_ratio=decrease:normalize_sar=true:upscaler=ewa_lanczos:downscaler=ewa_lanczos
libplacebo=fps=60:frame_mixer=mitchell_clamp
libplacebo=format=yuv420p:colorspace=bt470bg:color_primaries=bt709:color_trc=iec61966-2-1:range=pc
libplacebo=deband=true:deband_iterations=3:deband_radius=8:deband_threshold=6
ffmpeg ... -init_hw_device vulkan:llvmpipe ... -vf libplacebo=upscaler=none:downscaler=none:peak_detect=false
ffmpeg -export_side_data +film_grain ... -vf libplacebo=apply_filmgrain=true
ffmpeg -init_hw_device vulkan -hwaccel vaapi -hwaccel_output_format vaapi ... -vf libplacebo
Calculate the VMAF (Video Multi-Method Assessment Fusion) score for a reference/distorted pair of input videos.
The first input is the distorted video, and the second input is the reference video.
The obtained VMAF score is printed through the logging system.
It requires Netflix’s vmaf library (libvmaf) as a pre-requisite.
After installing the library it can be enabled using:
./configure --enable-libvmaf
.
The filter has following options:
A ‘|‘ delimited list of vmaf models. Each model can be configured with a number of parameters.
Default value: "version=vmaf_v0.6.1"
A ‘|‘ delimited list of features. Each feature can be configured with a number of parameters.
Set the file path to be used to store log files.
Set the format of the log file (xml, json, csv, or sub).
Set the pool method to be used for computing vmaf.
Options are min
, harmonic_mean
or mean
(default).
Set number of threads to be used when initializing libvmaf.
Default value: 0
, no threads.
Set frame subsampling interval to be used.
This filter also supports the framesync options.
ffmpeg -i distorted.mpg -i reference.mpg -lavfi libvmaf=log_path=output.xml -f null -
ffmpeg -i distorted.mpg -i reference.mpg -lavfi libvmaf='model=version=vmaf_v0.6.1\\:name=vmaf|version=vmaf_v0.6.1neg\\:name=vmaf_neg' -f null -
ffmpeg -i distorted.mpg -i reference.mpg -lavfi libvmaf='feature=name=psnr|name=ciede' -f null -
ffmpeg -i distorted.mpg -i reference.mkv -lavfi "[0:v]settb=AVTB,setpts=PTS-STARTPTS[main];[1:v]settb=AVTB,setpts=PTS-STARTPTS[ref];[main][ref]libvmaf=log_fmt=json:log_path=output.json" -f null -
This is the CUDA variant of the libvmaf filter. It only accepts CUDA frames.
It requires Netflix’s vmaf library (libvmaf) as a pre-requisite.
After installing the library it can be enabled using:
./configure --enable-nonfree --enable-ffnvcodec --enable-libvmaf
.
ffmpeg \ -hwaccel cuda -hwaccel_output_format cuda -codec:v av1_cuvid -i dis.obu \ -hwaccel cuda -hwaccel_output_format cuda -codec:v av1_cuvid -i ref.obu \ -filter_complex " [0:v]scale_cuda=format=yuv420p[ref]; \ [1:v]scale_cuda=format=yuv420p[dis]; \ [dis][ref]libvmaf_cuda=log_fmt=json:log_path=output.json " \ -f null -
Apply limited difference filter using second and optionally third video stream.
The filter accepts the following options:
Set the threshold to use when allowing certain differences between video streams. Any absolute difference value lower or exact than this threshold will pick pixel components from first video stream.
Set the elasticity of soft thresholding when processing video streams. This value multiplied with first one sets second threshold. Any absolute difference value greater or exact than second threshold will pick pixel components from second video stream. For values between those two threshold linear interpolation between first and second video stream will be used.
Enable the reference (third) video stream processing. By default is disabled. If set, this video stream will be used for calculating absolute difference with first video stream.
Specify which planes will be processed. Defaults to all available.
This filter supports the all above options as commands except option ‘reference’.
Limits the pixel components values to the specified range [min, max].
The filter accepts the following options:
Lower bound. Defaults to the lowest allowed value for the input.
Upper bound. Defaults to the highest allowed value for the input.
Specify which planes will be processed. Defaults to all available.
This filter supports the all above options as commands.
Loop video frames.
The filter accepts the following options:
Set the number of loops. Setting this value to -1 will result in infinite loops. Default is 0.
Set maximal size in number of frames. Default is 0.
Set first frame of loop. Default is 0.
Set the time of loop start in seconds.
Only used if option named start is set to -1
.
loop=loop=-1:size=1:start=0
loop=loop=10:size=1:start=0
loop=loop=5:size=10:start=0
Apply a 1D LUT to an input video.
The filter accepts the following options:
Set the 1D LUT file name.
Currently supported formats:
Iridas
cineSpace
Select interpolation mode.
Available values are:
Use values from the nearest defined point.
Interpolate values using the linear interpolation.
Interpolate values using the cosine interpolation.
Interpolate values using the cubic interpolation.
Interpolate values using the spline interpolation.
Apply a 3D LUT to an input video.
The filter accepts the following options:
Set the 3D LUT file name.
Currently supported formats:
AfterEffects
Iridas
DaVinci
Pandora
cineSpace
Select interpolation mode.
Available values are:
Use values from the nearest defined point.
Interpolate values using the 8 points defining a cube.
Interpolate values using a tetrahedron.
Interpolate values using a pyramid.
Interpolate values using a prism.
This filter supports the interp
option as commands.
Turn certain luma values into transparency.
The filter accepts the following options:
Set the luma which will be used as base for transparency.
Default value is 0
.
Set the range of luma values to be keyed out.
Default value is 0.01
.
Set the range of softness. Default value is 0
.
Use this to control gradual transition from zero to full transparency.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Compute a look-up table for binding each pixel component input value to an output value, and apply it to the input video.
lutyuv applies a lookup table to a YUV input video, lutrgb to an RGB input video.
These filters accept the following parameters:
set first pixel component expression
set second pixel component expression
set third pixel component expression
set fourth pixel component expression, corresponds to the alpha component
set red component expression
set green component expression
set blue component expression
alpha component expression
set Y/luma component expression
set U/Cb component expression
set V/Cr component expression
Each of them specifies the expression to use for computing the lookup table for the corresponding pixel component values.
The exact component associated to each of the c* options depends on the format in input.
The lut filter requires either YUV or RGB pixel formats in input, lutrgb requires RGB pixel formats in input, and lutyuv requires YUV.
The expressions can contain the following constants and functions:
The input width and height.
The input value for the pixel component.
The input value, clipped to the minval-maxval range.
The maximum value for the pixel component.
The minimum value for the pixel component.
The negated value for the pixel component value, clipped to the minval-maxval range; it corresponds to the expression "maxval-clipval+minval".
The computed value in val, clipped to the minval-maxval range.
The computed gamma correction value of the pixel component value, clipped to the minval-maxval range. It corresponds to the expression "pow((clipval-minval)/(maxval-minval)\,gamma)*(maxval-minval)+minval"
All expressions default to "clipval".
This filter supports same commands as options.
lutrgb="r=maxval+minval-val:g=maxval+minval-val:b=maxval+minval-val" lutyuv="y=maxval+minval-val:u=maxval+minval-val:v=maxval+minval-val"
The above is the same as:
lutrgb="r=negval:g=negval:b=negval" lutyuv="y=negval:u=negval:v=negval"
lutyuv=y=negval
lutyuv="u=128:v=128"
lutyuv="y=2*val"
lutrgb="g=0:b=0"
format=rgba,lutrgb=a="maxval-minval/2"
lutyuv=y=gammaval(0.5)
lutyuv=y='bitand(val, 128+64+32)'
lutyuv=u='(val-maxval/2)*2+maxval/2':v='(val-maxval/2)*2+maxval/2'
The lut2
filter takes two input streams and outputs one
stream.
The tlut2
(time lut2) filter takes two consecutive frames
from one single stream.
This filter accepts the following parameters:
set first pixel component expression
set second pixel component expression
set third pixel component expression
set fourth pixel component expression, corresponds to the alpha component
set output bit depth, only available for lut2
filter. By default is 0,
which means bit depth is automatically picked from first input format.
The lut2
filter also supports the framesync options.
Each of them specifies the expression to use for computing the lookup table for the corresponding pixel component values.
The exact component associated to each of the c* options depends on the format in inputs.
The expressions can contain the following constants:
The input width and height.
The first input value for the pixel component.
The second input value for the pixel component.
The first input video bit depth.
The second input video bit depth.
All expressions default to "x".
This filter supports the all above options as commands except option d
.
lut2='ifnot(x-y,0,pow(2,bdx)-1):ifnot(x-y,0,pow(2,bdx)-1):ifnot(x-y,0,pow(2,bdx)-1)'
lut2='ifnot(x-y,0,pow(2,bdx)-1):ifnot(x-y,pow(2,bdx-1),pow(2,bdx)-1):ifnot(x-y,pow(2,bdx-1),pow(2,bdx)-1)'
lut2='if(lt(x,y),0,if(gt(x,y),pow(2,bdx)-1,pow(2,bdx-1))):if(lt(x,y),0,if(gt(x,y),pow(2,bdx)-1,pow(2,bdx-1))):if(lt(x,y),0,if(gt(x,y),pow(2,bdx)-1,pow(2,bdx-1)))'
Clamp the first input stream with the second input and third input stream.
Returns the value of first stream to be between second input
stream - undershoot
and third input stream + overshoot
.
This filter accepts the following options:
Default value is 0
.
Default value is 0
.
Set which planes will be processed as bitmap, unprocessed planes will be copied from first stream. By default value 0xf, all planes will be processed.
This filter supports the all above options as commands.
Merge the second and third input stream into output stream using absolute differences between second input stream and first input stream and absolute difference between third input stream and first input stream. The picked value will be from second input stream if second absolute difference is greater than first one or from third input stream otherwise.
This filter accepts the following options:
Set which planes will be processed as bitmap, unprocessed planes will be copied from first stream. By default value 0xf, all planes will be processed.
This filter supports the all above options as commands.
Merge the first input stream with the second input stream using per pixel weights in the third input stream.
A value of 0 in the third stream pixel component means that pixel component from first stream is returned unchanged, while maximum value (eg. 255 for 8-bit videos) means that pixel component from second stream is returned unchanged. Intermediate values define the amount of merging between both input stream’s pixel components.
This filter accepts the following options:
Set which planes will be processed as bitmap, unprocessed planes will be copied from first stream. By default value 0xf, all planes will be processed.
This filter supports the all above options as commands.
Merge the second and third input stream into output stream using absolute differences between second input stream and first input stream and absolute difference between third input stream and first input stream. The picked value will be from second input stream if second absolute difference is less than first one or from third input stream otherwise.
This filter accepts the following options:
Set which planes will be processed as bitmap, unprocessed planes will be copied from first stream. By default value 0xf, all planes will be processed.
This filter supports the all above options as commands.
Pick pixels comparing absolute difference of two video streams with fixed threshold.
If absolute difference between pixel component of first and second video stream is equal or lower than user supplied threshold than pixel component from first video stream is picked, otherwise pixel component from second video stream is picked.
This filter accepts the following options:
Set threshold used when picking pixels from absolute difference from two input video streams.
Set which planes will be processed as bitmap, unprocessed planes will be copied from second stream. By default value 0xf, all planes will be processed.
Set mode of filter operation. Can be abs
or diff
.
Default is abs
.
This filter supports the all above options as commands.
Create mask from input video.
For example it is useful to create motion masks after tblend
filter.
This filter accepts the following options:
Set low threshold. Any pixel component lower or exact than this value will be set to 0.
Set high threshold. Any pixel component higher than this value will be set to max value allowed for current pixel format.
Set planes to filter, by default all available planes are filtered.
Fill all frame pixels with this value.
Set max average pixel value for frame. If sum of all pixel components is higher that this
average, output frame will be completely filled with value set by fill option.
Typically useful for scene changes when used in combination with tblend
filter.
This filter supports the all above options as commands.
Apply motion-compensation deinterlacing.
It needs one field per frame as input and must thus be used together with yadif=1/3 or equivalent.
This filter accepts the following options:
Set the deinterlacing mode.
It accepts one of the following values:
use iterative motion estimation
like ‘slow’, but use multiple reference frames.
Default value is ‘fast’.
Set the picture field parity assumed for the input video. It must be one of the following values:
assume top field first
assume bottom field first
Default value is ‘bff’.
Set per-block quantization parameter (QP) used by the internal encoder.
Higher values should result in a smoother motion vector field but less optimal individual vectors. Default value is 1.
Pick median pixel from certain rectangle defined by radius.
This filter accepts the following options:
Set horizontal radius size. Default value is 1
.
Allowed range is integer from 1 to 127.
Set which planes to process. Default is 15
, which is all available planes.
Set vertical radius size. Default value is 0
.
Allowed range is integer from 0 to 127.
If it is 0, value will be picked from horizontal radius
option.
Set median percentile. Default value is 0.5
.
Default value of 0.5
will pick always median values, while 0
will pick
minimum values, and 1
maximum values.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Merge color channel components from several video streams.
The filter accepts up to 4 input streams, and merge selected input planes to the output video.
This filter accepts the following options:
Set input to output plane mapping. Default is 0
.
The mappings is specified as a bitmap. It should be specified as a hexadecimal number in the form 0xAa[Bb[Cc[Dd]]]. ’Aa’ describes the mapping for the first plane of the output stream. ’A’ sets the number of the input stream to use (from 0 to 3), and ’a’ the plane number of the corresponding input to use (from 0 to 3). The rest of the mappings is similar, ’Bb’ describes the mapping for the output stream second plane, ’Cc’ describes the mapping for the output stream third plane and ’Dd’ describes the mapping for the output stream fourth plane.
Set output pixel format. Default is yuva444p
.
Set input to output stream mapping for output Nth plane. Default is 0
.
Set input to output plane mapping for output Nth plane. Default is 0
.
[a0][a1][a2]mergeplanes=0x001020:yuv444p
[a0][a1]mergeplanes=0x00010210:yuva444p
format=yuva444p,mergeplanes=0x03010200:yuva444p
format=yuv420p,mergeplanes=0x000201:yuv420p
format=rgb24,mergeplanes=0x000102:yuv444p
Estimate and export motion vectors using block matching algorithms. Motion vectors are stored in frame side data to be used by other filters.
This filter accepts the following options:
Specify the motion estimation method. Accepts one of the following values:
Exhaustive search algorithm.
Three step search algorithm.
Two dimensional logarithmic search algorithm.
New three step search algorithm.
Four step search algorithm.
Diamond search algorithm.
Hexagon-based search algorithm.
Enhanced predictive zonal search algorithm.
Uneven multi-hexagon search algorithm.
Default value is ‘esa’.
Macroblock size. Default 16
.
Search parameter. Default 7
.
Apply Midway Image Equalization effect using two video streams.
Midway Image Equalization adjusts a pair of images to have the same histogram, while maintaining their dynamics as much as possible. It’s useful for e.g. matching exposures from a pair of stereo cameras.
This filter has two inputs and one output, which must be of same pixel format, but may be of different sizes. The output of filter is first input adjusted with midway histogram of both inputs.
This filter accepts the following option:
Set which planes to process. Default is 15
, which is all available planes.
Convert the video to specified frame rate using motion interpolation.
This filter accepts the following options:
Specify the output frame rate. This can be rational e.g. 60000/1001
. Frames are dropped if fps is lower than source fps. Default 60
.
Motion interpolation mode. Following values are accepted:
Duplicate previous or next frame for interpolating new ones.
Blend source frames. Interpolated frame is mean of previous and next frames.
Motion compensated interpolation. Following options are effective when this mode is selected:
Motion compensation mode. Following values are accepted:
Overlapped block motion compensation.
Adaptive overlapped block motion compensation. Window weighting coefficients are controlled adaptively according to the reliabilities of the neighboring motion vectors to reduce oversmoothing.
Default mode is ‘obmc’.
Motion estimation mode. Following values are accepted:
Bidirectional motion estimation. Motion vectors are estimated for each source frame in both forward and backward directions.
Bilateral motion estimation. Motion vectors are estimated directly for interpolated frame.
Default mode is ‘bilat’.
The algorithm to be used for motion estimation. Following values are accepted:
Exhaustive search algorithm.
Three step search algorithm.
Two dimensional logarithmic search algorithm.
New three step search algorithm.
Four step search algorithm.
Diamond search algorithm.
Hexagon-based search algorithm.
Enhanced predictive zonal search algorithm.
Uneven multi-hexagon search algorithm.
Default algorithm is ‘epzs’.
Macroblock size. Default 16
.
Motion estimation search parameter. Default 32
.
Enable variable-size block motion compensation. Motion estimation is applied with smaller block sizes at object boundaries in order to make the them less blur. Default is 0
(disabled).
Scene change detection method. Scene change leads motion vectors to be in random direction. Scene change detection replace interpolated frames by duplicate ones. May not be needed for other modes. Following values are accepted:
Disable scene change detection.
Frame difference. Corresponding pixel values are compared and if it satisfies scd_threshold scene change is detected.
Default method is ‘fdiff’.
Scene change detection threshold. Default is 10.
.
Mix several video input streams into one video stream.
A description of the accepted options follows.
The number of inputs. If unspecified, it defaults to 2.
Specify weight of each input video stream as sequence. Each weight is separated by space. If number of weights is smaller than number of frames last specified weight will be used for all remaining unset weights.
Specify scale, if it is set it will be multiplied with sum of each weight multiplied with pixel values to give final destination pixel value. By default scale is auto scaled to sum of weights.
Set which planes to filter. Default is all. Allowed range is from 0 to 15.
Specify how end of stream is determined.
The duration of the longest input. (default)
The duration of the shortest input.
The duration of the first input.
This filter supports the following commands:
Syntax is same as option with same name.
Convert video to gray using custom color filter.
A description of the accepted options follows.
Set the chroma blue spot. Allowed range is from -1 to 1. Default value is 0.
Set the chroma red spot. Allowed range is from -1 to 1. Default value is 0.
Set the color filter size. Allowed range is from .1 to 10. Default value is 1.
Set the highlights strength. Allowed range is from 0 to 1. Default value is 0.
This filter supports the all above options as commands.
This filter allows to apply main morphological grayscale transforms, erode and dilate with arbitrary structures set in second input stream.
Unlike naive implementation and much slower performance in erosion
and dilation filters, when speed is critical morpho
filter
should be used instead.
A description of accepted options follows,
Set morphological transform to apply, can be:
Default is erode
.
Set planes to filter, by default all planes except alpha are filtered.
Set which structure video frames will be processed from second input stream, can be first or all. Default is all.
The morpho
filter also supports the framesync options.
This filter supports same commands as options.
Drop frames that do not differ greatly from the previous frame in order to reduce frame rate.
The main use of this filter is for very-low-bitrate encoding (e.g. streaming over dialup modem), but it could in theory be used for fixing movies that were inverse-telecined incorrectly.
A description of the accepted options follows.
Set the maximum number of consecutive frames which can be dropped (if positive), or the minimum interval between dropped frames (if negative). If the value is 0, the frame is dropped disregarding the number of previous sequentially dropped frames.
Default value is 0.
Set the maximum number of consecutive similar frames to ignore before to start dropping them. If the value is 0, the frame is dropped disregarding the number of previous sequentially similar frames.
Default value is 0.
Set the dropping threshold values.
Values for hi and lo are for 8x8 pixel blocks and represent actual pixel value differences, so a threshold of 64 corresponds to 1 unit of difference for each pixel, or the same spread out differently over the block.
A frame is a candidate for dropping if no 8x8 blocks differ by more than a threshold of hi, and if no more than frac blocks (1 meaning the whole image) differ by more than a threshold of lo.
Default value for hi is 64*12, default value for lo is 64*5, and default value for frac is 0.33.
Obtain the MSAD (Mean Sum of Absolute Differences) between two input videos.
This filter takes two input videos.
Both input videos must have the same resolution and pixel format for this filter to work correctly. Also it assumes that both inputs have the same number of frames, which are compared one by one.
The obtained per component, average, min and max MSAD is printed through the logging system.
The filter stores the calculated MSAD of each frame in frame metadata.
This filter also supports the framesync options.
In the below example the input file main.mpg being processed is compared with the reference file ref.mpg.
ffmpeg -i main.mpg -i ref.mpg -lavfi msad -f null -
Multiply first video stream pixels values with second video stream pixels values.
The filter accepts the following options:
Set the scale applied to second video stream. By default is 1
.
Allowed range is from 0
to 9
.
Set the offset applied to second video stream. By default is 0.5
.
Allowed range is from -1
to 1
.
Specify planes from input video stream that will be processed. By default all planes are processed.
This filter supports same commands as options.
Negate (invert) the input video.
It accepts the following option:
Set components to negate.
Available values for components are:
With value 1, it negates the alpha component, if present. Default value is 0.
Denoise frames using Non-Local Means algorithm.
Each pixel is adjusted by looking for other pixels with similar contexts. This context similarity is defined by comparing their surrounding patches of size pxp. Patches are searched in an area of rxr around the pixel.
Note that the research area defines centers for patches, which means some patches will be made of pixels outside that research area.
The filter accepts the following options.
Set denoising strength. Default is 1.0. Must be in range [1.0, 30.0].
Set patch size. Default is 7. Must be odd number in range [0, 99].
Same as p but for chroma planes.
The default value is 0 and means automatic.
Set research size. Default is 15. Must be odd number in range [0, 99].
Same as r but for chroma planes.
The default value is 0 and means automatic.
Deinterlace video using neural network edge directed interpolation.
This filter accepts the following options:
Mandatory option, without binary file filter can not work. Currently file can be found here: https://github.com/dubhater/vapoursynth-nnedi3/blob/master/src/nnedi3_weights.bin
Set which frames to deinterlace, by default it is all
.
Can be all
or interlaced
.
Set mode of operation.
Can be one of the following:
Use frame flags, both fields.
Use frame flags, single field.
Use top field only.
Use bottom field only.
Use both fields, top first.
Use both fields, bottom first.
Set which planes to process, by default filter process all frames.
Set size of local neighborhood around each pixel, used by the predictor neural network.
Can be one of the following:
Set the number of neurons in predictor neural network. Can be one of the following:
Controls the number of different neural network predictions that are blended
together to compute the final output value. Can be fast
, default or
slow
.
Set which set of weights to use in the predictor. Can be one of the following:
weights trained to minimize absolute error
weights trained to minimize squared error
Controls whether or not the prescreener neural network is used to decide which pixels should be processed by the predictor neural network and which can be handled by simple cubic interpolation. The prescreener is trained to know whether cubic interpolation will be sufficient for a pixel or whether it should be predicted by the predictor nn. The computational complexity of the prescreener nn is much less than that of the predictor nn. Since most pixels can be handled by cubic interpolation, using the prescreener generally results in much faster processing. The prescreener is pretty accurate, so the difference between using it and not using it is almost always unnoticeable.
Can be one of the following:
Default is new
.
This filter supports same commands as options, excluding weights option.
Force libavfilter not to use any of the specified pixel formats for the input to the next filter.
It accepts the following parameters:
A ’|’-separated list of pixel format names, such as pix_fmts=yuv420p|monow|rgb24".
noformat=pix_fmts=yuv420p,vflip
noformat=yuv420p|yuv444p|yuv410p
Add noise on video input frame.
The filter accepts the following options:
Set noise seed for specific pixel component or all pixel components in case
of all_seed. Default value is 123457
.
Set noise strength for specific pixel component or all pixel components in case
all_strength. Default value is 0
. Allowed range is [0, 100].
Set pixel component flags or set flags for all components if all_flags. Available values for component flags are:
averaged temporal noise (smoother)
mix random noise with a (semi)regular pattern
temporal noise (noise pattern changes between frames)
uniform noise (gaussian otherwise)
Add temporal and uniform noise to input video:
noise=alls=20:allf=t+u
Normalize RGB video (aka histogram stretching, contrast stretching). See: https://en.wikipedia.org/wiki/Normalization_(image_processing)
For each channel of each frame, the filter computes the input range and maps it linearly to the user-specified output range. The output range defaults to the full dynamic range from pure black to pure white.
Temporal smoothing can be used on the input range to reduce flickering (rapid changes in brightness) caused when small dark or bright objects enter or leave the scene. This is similar to the auto-exposure (automatic gain control) on a video camera, and, like a video camera, it may cause a period of over- or under-exposure of the video.
The R,G,B channels can be normalized independently, which may cause some color shifting, or linked together as a single channel, which prevents color shifting. Linked normalization preserves hue. Independent normalization does not, so it can be used to remove some color casts. Independent and linked normalization can be combined in any ratio.
The normalize filter accepts the following options:
Colors which define the output range. The minimum input value is mapped to the blackpt. The maximum input value is mapped to the whitept. The defaults are black and white respectively. Specifying white for blackpt and black for whitept will give color-inverted, normalized video. Shades of grey can be used to reduce the dynamic range (contrast). Specifying saturated colors here can create some interesting effects.
The number of previous frames to use for temporal smoothing. The input range of each channel is smoothed using a rolling average over the current frame and the smoothing previous frames. The default is 0 (no temporal smoothing).
Controls the ratio of independent (color shifting) channel normalization to linked (color preserving) normalization. 0.0 is fully linked, 1.0 is fully independent. Defaults to 1.0 (fully independent).
Overall strength of the filter. 1.0 is full strength. 0.0 is a rather expensive no-op. Defaults to 1.0 (full strength).
This filter supports same commands as options, excluding smoothing option. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Stretch video contrast to use the full dynamic range, with no temporal smoothing; may flicker depending on the source content:
normalize=blackpt=black:whitept=white:smoothing=0
As above, but with 50 frames of temporal smoothing; flicker should be reduced, depending on the source content:
normalize=blackpt=black:whitept=white:smoothing=50
As above, but with hue-preserving linked channel normalization:
normalize=blackpt=black:whitept=white:smoothing=50:independence=0
As above, but with half strength:
normalize=blackpt=black:whitept=white:smoothing=50:independence=0:strength=0.5
Map the darkest input color to red, the brightest input color to cyan:
normalize=blackpt=red:whitept=cyan
Pass the video source unchanged to the output.
Optical Character Recognition
This filter uses Tesseract for optical character recognition. To enable
compilation of this filter, you need to configure FFmpeg with
--enable-libtesseract
.
It accepts the following options:
Set datapath to tesseract data. Default is to use whatever was set at installation.
Set language, default is "eng".
Set character whitelist.
Set character blacklist.
The filter exports recognized text as the frame metadata lavfi.ocr.text
.
The filter exports confidence of recognized words as the frame metadata lavfi.ocr.confidence
.
Apply a video transform using libopencv.
To enable this filter, install the libopencv library and headers and
configure FFmpeg with --enable-libopencv
.
It accepts the following parameters:
The name of the libopencv filter to apply.
The parameters to pass to the libopencv filter. If not specified, the default values are assumed.
Refer to the official libopencv documentation for more precise information: http://docs.opencv.org/master/modules/imgproc/doc/filtering.html
Several libopencv filters are supported; see the following subsections.
Dilate an image by using a specific structuring element.
It corresponds to the libopencv function cvDilate
.
It accepts the parameters: struct_el|nb_iterations.
struct_el represents a structuring element, and has the syntax: colsxrows+anchor_xxanchor_y/shape
cols and rows represent the number of columns and rows of the structuring element, anchor_x and anchor_y the anchor point, and shape the shape for the structuring element. shape must be "rect", "cross", "ellipse", or "custom".
If the value for shape is "custom", it must be followed by a string of the form "=filename". The file with name filename is assumed to represent a binary image, with each printable character corresponding to a bright pixel. When a custom shape is used, cols and rows are ignored, the number or columns and rows of the read file are assumed instead.
The default value for struct_el is "3x3+0x0/rect".
nb_iterations specifies the number of times the transform is applied to the image, and defaults to 1.
Some examples:
# Use the default values ocv=dilate # Dilate using a structuring element with a 5x5 cross, iterating two times ocv=filter_name=dilate:filter_params=5x5+2x2/cross|2 # Read the shape from the file diamond.shape, iterating two times. # The file diamond.shape may contain a pattern of characters like this # * # *** # ***** # *** # * # The specified columns and rows are ignored # but the anchor point coordinates are not ocv=dilate:0x0+2x2/custom=diamond.shape|2
Erode an image by using a specific structuring element.
It corresponds to the libopencv function cvErode
.
It accepts the parameters: struct_el:nb_iterations, with the same syntax and semantics as the dilate filter.
Smooth the input video.
The filter takes the following parameters: type|param1|param2|param3|param4.
type is the type of smooth filter to apply, and must be one of the following values: "blur", "blur_no_scale", "median", "gaussian", or "bilateral". The default value is "gaussian".
The meaning of param1, param2, param3, and param4 depends on the smooth type. param1 and param2 accept integer positive values or 0. param3 and param4 accept floating point values.
The default value for param1 is 3. The default value for the other parameters is 0.
These parameters correspond to the parameters assigned to the
libopencv function cvSmooth
.
2D Video Oscilloscope.
Useful to measure spatial impulse, step responses, chroma delays, etc.
It accepts the following parameters:
Set scope center x position.
Set scope center y position.
Set scope size, relative to frame diagonal.
Set scope tilt/rotation.
Set trace opacity.
Set trace center x position.
Set trace center y position.
Set trace width, relative to width of frame.
Set trace height, relative to height of frame.
Set which components to trace. By default it traces first three components.
Draw trace grid. By default is enabled.
Draw some statistics. By default is enabled.
Draw scope. By default is enabled.
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
oscilloscope=x=0.5:y=0:s=1
oscilloscope=x=0.5:y=1:s=1
oscilloscope=x=0.5:y=5/1080:s=1
oscilloscope=x=1:y=0.5:s=1:t=1
Overlay one video on top of another.
It takes two inputs and has one output. The first input is the "main" video on which the second input is overlaid.
It accepts the following parameters:
A description of the accepted options follows.
Set the expression for the x and y coordinates of the overlaid video on the main video. Default value is "0" for both expressions. In case the expression is invalid, it is set to a huge value (meaning that the overlay will not be displayed within the output visible area).
See framesync.
Set when the expressions for x, and y are evaluated.
It accepts the following values:
only evaluate expressions once during the filter initialization or when a command is processed
evaluate expressions for each incoming frame
Default value is ‘frame’.
See framesync.
Set the format for the output video.
It accepts the following values:
force YUV 4:2:0 8-bit planar output
force YUV 4:2:0 10-bit planar output
force YUV 4:2:2 8-bit planar output
force YUV 4:2:2 10-bit planar output
force YUV 4:4:4 8-bit planar output
force YUV 4:4:4 10-bit planar output
force RGB 8-bit packed output
force RGB 8-bit planar output
automatically pick format
Default value is ‘yuv420’.
See framesync.
Set format of alpha of the overlaid video, it can be straight or premultiplied. Default is straight.
The x, and y expressions can contain the following parameters.
The main input width and height.
The overlay input width and height.
The computed values for x and y. They are evaluated for each new frame.
horizontal and vertical chroma subsample values of the output format. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
the number of input frame, starting from 0
the position in the file of the input frame, NAN if unknown; deprecated, do not use
The timestamp, expressed in seconds. It’s NAN if the input timestamp is unknown.
This filter also supports the framesync options.
Note that the n, t variables are available only when evaluation is done per frame, and will evaluate to NAN when eval is set to ‘init’.
Be aware that frames are taken from each input video in timestamp order, hence, if their initial timestamps differ, it is a good idea to pass the two inputs through a setpts=PTS-STARTPTS filter to have them begin in the same zero timestamp, as the example for the movie filter does.
You can chain together more overlays but you should test the efficiency of such approach.
This filter supports the following commands:
Modify the x and y of the overlay input. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
overlay=main_w-overlay_w-10:main_h-overlay_h-10
Using named options the example above becomes:
overlay=x=main_w-overlay_w-10:y=main_h-overlay_h-10
ffmpeg
tool with the -filter_complex
option:
ffmpeg -i input -i logo -filter_complex 'overlay=10:main_h-overlay_h-10' output
ffmpeg
tool:
ffmpeg -i input -i logo1 -i logo2 -filter_complex 'overlay=x=10:y=H-h-10,overlay=x=W-w-10:y=H-h-10' output
WxH
must specify the size of the main input to the overlay filter:
color=color=red@.3:size=WxH [over]; [in][over] overlay [out]
ffplay
tool:
ffplay input.avi -vf 'split[a][b]; [a]pad=iw*2:ih[src]; [b]deshake[filt]; [src][filt]overlay=w'
The above command is the same as:
ffplay input.avi -vf 'split[b], pad=iw*2[src], [b]deshake, [src]overlay=w'
overlay=x='if(gte(t,2), -w+(t-2)*20, NAN)':y=0
ffmpeg -i left.avi -i right.avi -filter_complex " nullsrc=size=200x100 [background]; [0:v] setpts=PTS-STARTPTS, scale=100x100 [left]; [1:v] setpts=PTS-STARTPTS, scale=100x100 [right]; [background][left] overlay=shortest=1 [background+left]; [background+left][right] overlay=shortest=1:x=100 [left+right] "
ffmpeg -i test.avi -codec:v:0 wmv2 -ar 11025 -b:v 9000k -vf '[in]split[split_main][split_delogo];[split_delogo]trim=start=360:end=371,delogo=0:0:640:480[delogoed];[split_main][delogoed]overlay=eof_action=pass[out]' masked.avi
nullsrc=s=200x200 [bg]; testsrc=s=100x100, split=4 [in0][in1][in2][in3]; [in0] lutrgb=r=0, [bg] overlay=0:0 [mid0]; [in1] lutrgb=g=0, [mid0] overlay=100:0 [mid1]; [in2] lutrgb=b=0, [mid1] overlay=0:100 [mid2]; [in3] null, [mid2] overlay=100:100 [out0]
Overlay one video on top of another.
This is the CUDA variant of the overlay filter. It only accepts CUDA frames. The underlying input pixel formats have to match.
It takes two inputs and has one output. The first input is the "main" video on which the second input is overlaid.
It accepts the following parameters:
Set expressions for the x and y coordinates of the overlaid video on the main video.
They can contain the following parameters:
The main input width and height.
The overlay input width and height.
The computed values for x and y. They are evaluated for each new frame.
The ordinal index of the main input frame, starting from 0.
The byte offset position in the file of the main input frame, NAN if unknown. Deprecated, do not use.
The timestamp of the main input frame, expressed in seconds, NAN if unknown.
Default value is "0" for both expressions.
Set when the expressions for x and y are evaluated.
It accepts the following values:
Evaluate expressions once during filter initialization or when a command is processed.
Evaluate expressions for each incoming frame
Default value is frame.
See framesync.
See framesync.
See framesync.
This filter also supports the framesync options.
Apply Overcomplete Wavelet denoiser.
The filter accepts the following options:
Set depth.
Larger depth values will denoise lower frequency components more, but slow down filtering.
Must be an int in the range 8-16, default is 8
.
Set luma strength.
Must be a double value in the range 0-1000, default is 1.0
.
Set chroma strength.
Must be a double value in the range 0-1000, default is 1.0
.
Add paddings to the input image, and place the original input at the provided x, y coordinates.
It accepts the following parameters:
Specify an expression for the size of the output image with the paddings added. If the value for width or height is 0, the corresponding input size is used for the output.
The width expression can reference the value set by the height expression, and vice versa.
The default value of width and height is 0.
Specify the offsets to place the input image at within the padded area, with respect to the top/left border of the output image.
The x expression can reference the value set by the y expression, and vice versa.
The default value of x and y is 0.
If x or y evaluate to a negative number, they’ll be changed so the input image is centered on the padded area.
Specify the color of the padded area. For the syntax of this option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual.
The default value of color is "black".
Specify when to evaluate width, height, x and y expression.
It accepts the following values:
Only evaluate expressions once during the filter initialization or when a command is processed.
Evaluate expressions for each incoming frame.
Default value is ‘init’.
Pad to aspect instead to a resolution.
The value for the width, height, x, and y options are expressions containing the following constants:
The input video width and height.
These are the same as in_w and in_h.
The output width and height (the size of the padded area), as specified by the width and height expressions.
These are the same as out_w and out_h.
The x and y offsets as specified by the x and y expressions, or NAN if not yet specified.
same as iw / ih
input sample aspect ratio
input display aspect ratio, it is the same as (iw / ih) * sar
The horizontal and vertical chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
pad=640:480:0:40:violet
The example above is equivalent to the following command:
pad=width=640:height=480:x=0:y=40:color=violet
pad="3/2*iw:3/2*ih:(ow-iw)/2:(oh-ih)/2"
pad="max(iw\,ih):ow:(ow-iw)/2:(oh-ih)/2"
pad="ih*16/9:ih:(ow-iw)/2:(oh-ih)/2"
(ih * X / ih) * sar = output_dar X = output_dar / sar
Thus the previous example needs to be modified to:
pad="ih*16/9/sar:ih:(ow-iw)/2:(oh-ih)/2"
pad="2*iw:2*ih:ow-iw:oh-ih"
Generate one palette for a whole video stream.
It accepts the following options:
Set the maximum number of colors to quantize in the palette. Note: the palette will still contain 256 colors; the unused palette entries will be black.
Create a palette of 255 colors maximum and reserve the last one for transparency. Reserving the transparency color is useful for GIF optimization. If not set, the maximum of colors in the palette will be 256. You probably want to disable this option for a standalone image. Set by default.
Set the color that will be used as background for transparency.
Set statistics mode.
It accepts the following values:
Compute full frame histograms.
Compute histograms only for the part that differs from previous frame. This might be relevant to give more importance to the moving part of your input if the background is static.
Compute new histogram for each frame.
Default value is full.
The filter also exports the frame metadata lavfi.color_quant_ratio
(nb_color_in / nb_color_out
) which you can use to evaluate the degree of
color quantization of the palette. This information is also visible at
info logging level.
ffmpeg
:
ffmpeg -i input.mkv -vf palettegen palette.png
Use a palette to downsample an input video stream.
The filter takes two inputs: one video stream and a palette. The palette must be a 256 pixels image.
It accepts the following options:
Select dithering mode. Available algorithms are:
Ordered 8x8 bayer dithering (deterministic)
Dithering as defined by Paul Heckbert in 1982 (simple error diffusion). Note: this dithering is sometimes considered "wrong" and is included as a reference.
Floyd and Steingberg dithering (error diffusion)
Frankie Sierra dithering v2 (error diffusion)
Frankie Sierra dithering v2 "Lite" (error diffusion)
Frankie Sierra dithering v3 (error diffusion)
Burkes dithering (error diffusion)
Atkinson dithering by Bill Atkinson at Apple Computer (error diffusion)
Disable dithering.
Default is sierra2_4a.
When bayer dithering is selected, this option defines the scale of the pattern (how much the crosshatch pattern is visible). A low value means more visible pattern for less banding, and higher value means less visible pattern at the cost of more banding.
The option must be an integer value in the range [0,5]. Default is 2.
If set, define the zone to process
Only the changing rectangle will be reprocessed. This is similar to GIF cropping/offsetting compression mechanism. This option can be useful for speed if only a part of the image is changing, and has use cases such as limiting the scope of the error diffusal dither to the rectangle that bounds the moving scene (it leads to more deterministic output if the scene doesn’t change much, and as a result less moving noise and better GIF compression).
Default is none.
Take new palette for each output frame.
Sets the alpha threshold for transparency. Alpha values above this threshold will be treated as completely opaque, and values below this threshold will be treated as completely transparent.
The option must be an integer value in the range [0,255]. Default is 128.
ffmpeg
:
ffmpeg -i input.mkv -i palette.png -lavfi paletteuse output.gif
Correct perspective of video not recorded perpendicular to the screen.
A description of the accepted parameters follows.
Set coordinates expression for top left, top right, bottom left and bottom right corners.
Default values are 0:0:W:0:0:H:W:H
with which perspective will remain unchanged.
If the sense
option is set to source
, then the specified points will be sent
to the corners of the destination. If the sense
option is set to destination
,
then the corners of the source will be sent to the specified coordinates.
The expressions can use the following variables:
the width and height of video frame.
Input frame count.
Output frame count.
Set interpolation for perspective correction.
It accepts the following values:
Default value is ‘linear’.
Set interpretation of coordinate options.
It accepts the following values:
Send point in the source specified by the given coordinates to the corners of the destination.
Send the corners of the source to the point in the destination specified by the given coordinates.
Default value is ‘source’.
Set when the expressions for coordinates x0,y0,...x3,y3 are evaluated.
It accepts the following values:
only evaluate expressions once during the filter initialization or when a command is processed
evaluate expressions for each incoming frame
Default value is ‘init’.
Delay interlaced video by one field time so that the field order changes.
The intended use is to fix PAL movies that have been captured with the opposite field order to the film-to-video transfer.
A description of the accepted parameters follows.
Set phase mode.
It accepts the following values:
Capture field order top-first, transfer bottom-first. Filter will delay the bottom field.
Capture field order bottom-first, transfer top-first. Filter will delay the top field.
Capture and transfer with the same field order. This mode only exists for the documentation of the other options to refer to, but if you actually select it, the filter will faithfully do nothing.
Capture field order determined automatically by field flags, transfer opposite. Filter selects among ‘t’ and ‘b’ modes on a frame by frame basis using field flags. If no field information is available, then this works just like ‘u’.
Capture unknown or varying, transfer opposite. Filter selects among ‘t’ and ‘b’ on a frame by frame basis by analyzing the images and selecting the alternative that produces best match between the fields.
Capture top-first, transfer unknown or varying. Filter selects among ‘t’ and ‘p’ using image analysis.
Capture bottom-first, transfer unknown or varying. Filter selects among ‘b’ and ‘p’ using image analysis.
Capture determined by field flags, transfer unknown or varying. Filter selects among ‘t’, ‘b’ and ‘p’ using field flags and image analysis. If no field information is available, then this works just like ‘U’. This is the default mode.
Both capture and transfer unknown or varying. Filter selects among ‘t’, ‘b’ and ‘p’ using image analysis only.
This filter supports the all above options as commands.
Reduce various flashes in video, so to help users with epilepsy.
It accepts the following options:
Set how many frames to use when filtering. Default is 30.
Set detection threshold factor. Default is 1. Lower is stricter.
Set how many pixels to skip when sampling frames. Default is 1. Allowed range is from 1 to 1024.
Leave frames unchanged. Default is disabled.
Pixel format descriptor test filter, mainly useful for internal testing. The output video should be equal to the input video.
For example:
format=monow, pixdesctest
can be used to test the monowhite pixel format descriptor definition.
Apply pixelization to video stream.
The filter accepts the following options:
Set block dimensions that will be used for pixelization.
Default value is 16
.
Set the mode of pixelization used.
Possible values are:
Default value is avg
.
Set what planes to filter. Default is to filter all planes.
This filter supports all options as commands.
Display sample values of color channels. Mainly useful for checking color and levels. Minimum supported resolution is 640x480.
The filters accept the following options:
Set scope X position, relative offset on X axis.
Set scope Y position, relative offset on Y axis.
Set scope width.
Set scope height.
Set window opacity. This window also holds statistics about pixel area.
Set window X position, relative offset on X axis.
Set window Y position, relative offset on Y axis.
This filter supports same commands as options.
Enable the specified chain of postprocessing subfilters using libpostproc. This
library should be automatically selected with a GPL build (--enable-gpl
).
Subfilters must be separated by ’/’ and can be disabled by prepending a ’-’.
Each subfilter and some options have a short and a long name that can be used
interchangeably, i.e. dr/dering are the same.
The filters accept the following options:
Set postprocessing subfilters string.
All subfilters share common options to determine their scope:
Honor the quality commands for this subfilter.
Do chrominance filtering, too (default).
Do luma filtering only (no chrominance).
Do chrominance filtering only (no luma).
These options can be appended after the subfilter name, separated by a ’|’.
Available subfilters are:
Horizontal deblocking filter
Difference factor where higher values mean more deblocking (default: 32
).
Flatness threshold where lower values mean more deblocking (default: 39
).
Vertical deblocking filter
Difference factor where higher values mean more deblocking (default: 32
).
Flatness threshold where lower values mean more deblocking (default: 39
).
Accurate horizontal deblocking filter
Difference factor where higher values mean more deblocking (default: 32
).
Flatness threshold where lower values mean more deblocking (default: 39
).
Accurate vertical deblocking filter
Difference factor where higher values mean more deblocking (default: 32
).
Flatness threshold where lower values mean more deblocking (default: 39
).
The horizontal and vertical deblocking filters share the difference and flatness values so you cannot set different horizontal and vertical thresholds.
Experimental horizontal deblocking filter
Experimental vertical deblocking filter
Deringing filter
larger -> stronger filtering
larger -> stronger filtering
larger -> stronger filtering
Stretch luma to 0-255
.
Linear blend deinterlacing filter that deinterlaces the given block by
filtering all lines with a (1 2 1)
filter.
Linear interpolating deinterlacing filter that deinterlaces the given block by linearly interpolating every second line.
Cubic interpolating deinterlacing filter deinterlaces the given block by cubically interpolating every second line.
Median deinterlacing filter that deinterlaces the given block by applying a median filter to every second line.
FFmpeg deinterlacing filter that deinterlaces the given block by filtering every
second line with a (-1 4 2 4 -1)
filter.
Vertically applied FIR lowpass deinterlacing filter that deinterlaces the given
block by filtering all lines with a (-1 2 6 2 -1)
filter.
Overrides the quantizer table from the input with the constant quantizer you specify.
Quantizer to use
Default pp filter combination (hb|a,vb|a,dr|a
)
Fast pp filter combination (h1|a,v1|a,dr|a
)
High quality pp filter combination (ha|a|128|7,va|a,dr|a
)
pp=hb/vb/dr/al
pp=de/-al
pp=default/tmpnoise|1|2|3
pp=hb|y/vb|a
Apply Postprocessing filter 7. It is variant of the spp filter, similar to spp = 6 with 7 point DCT, where only the center sample is used after IDCT.
The filter accepts the following options:
Force a constant quantization parameter. It accepts an integer in range 0 to 63. If not set, the filter will use the QP from the video stream (if available).
Set thresholding mode. Available modes are:
Set hard thresholding.
Set soft thresholding (better de-ringing effect, but likely blurrier).
Set medium thresholding (good results, default).
Apply alpha premultiply effect to input video stream using first plane of second stream as alpha.
Both streams must have same dimensions and same pixel format.
The filter accepts the following option:
Set which planes will be processed, unprocessed planes will be copied. By default value 0xf, all planes will be processed.
Do not require 2nd input for processing, instead use alpha plane from input stream.
Apply prewitt operator to input video stream.
The filter accepts the following option:
Set which planes will be processed, unprocessed planes will be copied. By default value 0xf, all planes will be processed.
Set value which will be multiplied with filtered result.
Set value which will be added to filtered result.
This filter supports the all above options as commands.
Alter frame colors in video with pseudocolors.
This filter accepts the following options:
set pixel first component expression
set pixel second component expression
set pixel third component expression
set pixel fourth component expression, corresponds to the alpha component
set component to use as base for altering colors
Pick one of built-in LUTs. By default is set to none.
Available LUTs:
Set opacity of output colors. Allowed range is from 0 to 1. Default value is set to 1.
Each of the expression options specifies the expression to use for computing the lookup table for the corresponding pixel component values.
The expressions can contain the following constants and functions:
The input width and height.
The input value for the pixel component.
The minimum allowed component value.
The maximum allowed component value.
All expressions default to "val".
This filter supports the all above options as commands.
pseudocolor="'if(between(val,ymax,amax),lerp(ymin,ymax,(val-ymax)/(amax-ymax)),-1):if(between(val,ymax,amax),lerp(umax,umin,(val-ymax)/(amax-ymax)),-1):if(between(val,ymax,amax),lerp(vmin,vmax,(val-ymax)/(amax-ymax)),-1):-1'"
Obtain the average, maximum and minimum PSNR (Peak Signal to Noise Ratio) between two input videos.
This filter takes in input two input videos, the first input is considered the "main" source and is passed unchanged to the output. The second input is used as a "reference" video for computing the PSNR.
Both video inputs must have the same resolution and pixel format for this filter to work correctly. Also it assumes that both inputs have the same number of frames, which are compared one by one.
The obtained average PSNR is printed through the logging system.
The filter stores the accumulated MSE (mean squared error) of each frame, and at the end of the processing it is averaged across all frames equally, and the following formula is applied to obtain the PSNR:
PSNR = 10*log10(MAX^2/MSE)
Where MAX is the average of the maximum values of each component of the image.
The description of the accepted parameters follows.
If specified the filter will use the named file to save the PSNR of each individual frame. When filename equals "-" the data is sent to standard output.
Specifies which version of the stats file format to use. Details of each format are written below. Default value is 1.
Determines whether the max value is output to the stats log. Default value is 0. Requires stats_version >= 2. If this is set and stats_version < 2, the filter will return an error.
This filter also supports the framesync options.
The file printed if stats_file is selected, contains a sequence of key/value pairs of the form key:value for each compared couple of frames.
If a stats_version greater than 1 is specified, a header line precedes the list of per-frame-pair stats, with key value pairs following the frame format with the following parameters:
The version of the log file format. Will match stats_version.
A comma separated list of the per-frame-pair parameters included in the log.
A description of each shown per-frame-pair parameter follows:
sequential number of the input frame, starting from 1
Mean Square Error pixel-by-pixel average difference of the compared frames, averaged over all the image components.
Mean Square Error pixel-by-pixel average difference of the compared frames for the component specified by the suffix.
Peak Signal to Noise ratio of the compared frames for the component specified by the suffix.
Maximum allowed value for each channel, and average over all channels.
movie=ref_movie.mpg, setpts=PTS-STARTPTS [main]; [main][ref] psnr="stats_file=stats.log" [out]
On this example the input file being processed is compared with the reference file ref_movie.mpg. The PSNR of each individual frame is stored in stats.log.
ffmpeg -i main.mpg -i ref.mkv -lavfi "[0:v]settb=AVTB,setpts=PTS-STARTPTS[main];[1:v]settb=AVTB,setpts=PTS-STARTPTS[ref];[main][ref]psnr" -f null -
Pulldown reversal (inverse telecine) filter, capable of handling mixed hard-telecine, 24000/1001 fps progressive, and 30000/1001 fps progressive content.
The pullup filter is designed to take advantage of future context in making its decisions. This filter is stateless in the sense that it does not lock onto a pattern to follow, but it instead looks forward to the following fields in order to identify matches and rebuild progressive frames.
To produce content with an even framerate, insert the fps filter after
pullup, use fps=24000/1001
if the input frame rate is 29.97fps,
fps=24
for 30fps and the (rare) telecined 25fps input.
The filter accepts the following options:
These options set the amount of "junk" to ignore at the left, right, top, and bottom of the image, respectively. Left and right are in units of 8 pixels, while top and bottom are in units of 2 lines. The default is 8 pixels on each side.
Set the strict breaks. Setting this option to 1 will reduce the chances of
filter generating an occasional mismatched frame, but it may also cause an
excessive number of frames to be dropped during high motion sequences.
Conversely, setting it to -1 will make filter match fields more easily.
This may help processing of video where there is slight blurring between
the fields, but may also cause there to be interlaced frames in the output.
Default value is 0
.
Set the metric plane to use. It accepts the following values:
Use luma plane.
Use chroma blue plane.
Use chroma red plane.
This option may be set to use chroma plane instead of the default luma plane for doing filter’s computations. This may improve accuracy on very clean source material, but more likely will decrease accuracy, especially if there is chroma noise (rainbow effect) or any grayscale video. The main purpose of setting mp to a chroma plane is to reduce CPU load and make pullup usable in realtime on slow machines.
For best results (without duplicated frames in the output file) it is necessary to change the output frame rate. For example, to inverse telecine NTSC input:
ffmpeg -i input -vf pullup -r 24000/1001 ...
Change video quantization parameters (QP).
The filter accepts the following option:
Set expression for quantization parameter.
The expression is evaluated through the eval API and can contain, among others, the following constants:
1 if index is not 129, 0 otherwise.
Sequential index starting from -129 to 128.
qp=2+2*sin(PI*qp)
Flush video frames from internal cache of frames into a random order. No frame is discarded. Inspired by frei0r nervous filter.
Set size in number of frames of internal cache, in range from 2
to
512
. Default is 30
.
Set seed for random number generator, must be an integer included between
0
and UINT32_MAX
. If not specified, or if explicitly set to
less than 0
, the filter will try to use a good random seed on a
best effort basis.
Read closed captioning (EIA-608) information from the top lines of a video frame.
This filter adds frame metadata for lavfi.readeia608.X.cc
and
lavfi.readeia608.X.line
, where X
is the number of the identified line
with EIA-608 data (starting from 0). A description of each metadata value follows:
The two bytes stored as EIA-608 data (printed in hexadecimal).
The number of the line on which the EIA-608 data was identified and read.
This filter accepts the following options:
Set the line to start scanning for EIA-608 data. Default is 0
.
Set the line to end scanning for EIA-608 data. Default is 29
.
Set the ratio of width reserved for sync code detection.
Default is 0.27
. Allowed range is [0.1 - 0.7]
.
Enable checking the parity bit. In the event of a parity error, the filter will output
0x00
for that character. Default is false.
Lowpass lines prior to further processing. Default is enabled.
This filter supports the all above options as commands.
ffprobe -f lavfi -i movie=captioned_video.mov,readeia608 -show_entries frame=pts_time:frame_tags=lavfi.readeia608.0.cc,lavfi.readeia608.1.cc -of csv
Read vertical interval timecode (VITC) information from the top lines of a video frame.
The filter adds frame metadata key lavfi.readvitc.tc_str
with the
timecode value, if a valid timecode has been detected. Further metadata key
lavfi.readvitc.found
is set to 0/1 depending on whether
timecode data has been found or not.
This filter accepts the following options:
Set the maximum number of lines to scan for VITC data. If the value is set to
-1
the full video frame is scanned. Default is 45
.
Set the luma threshold for black. Accepts float numbers in the range [0.0,1.0],
default value is 0.2
. The value must be equal or less than thr_w
.
Set the luma threshold for white. Accepts float numbers in the range [0.0,1.0],
default value is 0.6
. The value must be equal or greater than thr_b
.
--:--:--:--
as a placeholder:
ffmpeg -i input.avi -filter:v 'readvitc,drawtext=fontfile=FreeMono.ttf:text=%{metadata\\:lavfi.readvitc.tc_str\\:--\\\\\\:--\\\\\\:--\\\\\\:--}:x=(w-tw)/2:y=400-ascent'
Remap pixels using 2nd: Xmap and 3rd: Ymap input video stream.
Destination pixel at position (X, Y) will be picked from source (x, y) position where x = Xmap(X, Y) and y = Ymap(X, Y). If mapping values are out of range, zero value for pixel will be used for destination pixel.
Xmap and Ymap input video streams must be of same dimensions. Output video stream will have Xmap/Ymap video stream dimensions. Xmap and Ymap input video streams are 16bit depth, single channel.
Specify pixel format of output from this filter. Can be color
or gray
.
Default is color
.
Specify the color of the unmapped pixels. For the syntax of this option,
check the (ffmpeg-utils)"Color" section in the ffmpeg-utils
manual. Default color is black
.
The removegrain filter is a spatial denoiser for progressive video.
Set mode for the first plane.
Set mode for the second plane.
Set mode for the third plane.
Set mode for the fourth plane.
Range of mode is from 0 to 24. Description of each mode follows:
Leave input plane unchanged. Default.
Clips the pixel with the minimum and maximum of the 8 neighbour pixels.
Clips the pixel with the second minimum and maximum of the 8 neighbour pixels.
Clips the pixel with the third minimum and maximum of the 8 neighbour pixels.
Clips the pixel with the fourth minimum and maximum of the 8 neighbour pixels. This is equivalent to a median filter.
Line-sensitive clipping giving the minimal change.
Line-sensitive clipping, intermediate.
Line-sensitive clipping, intermediate.
Line-sensitive clipping, intermediate.
Line-sensitive clipping on a line where the neighbours pixels are the closest.
Replaces the target pixel with the closest neighbour.
[1 2 1] horizontal and vertical kernel blur.
Same as mode 11.
Bob mode, interpolates top field from the line where the neighbours pixels are the closest.
Bob mode, interpolates bottom field from the line where the neighbours pixels are the closest.
Bob mode, interpolates top field. Same as 13 but with a more complicated interpolation formula.
Bob mode, interpolates bottom field. Same as 14 but with a more complicated interpolation formula.
Clips the pixel with the minimum and maximum of respectively the maximum and minimum of each pair of opposite neighbour pixels.
Line-sensitive clipping using opposite neighbours whose greatest distance from the current pixel is minimal.
Replaces the pixel with the average of its 8 neighbours.
Averages the 9 pixels ([1 1 1] horizontal and vertical blur).
Clips pixels using the averages of opposite neighbour.
Same as mode 21 but simpler and faster.
Small edge and halo removal, but reputed useless.
Similar as 23.
Suppress a TV station logo, using an image file to determine which pixels comprise the logo. It works by filling in the pixels that comprise the logo with neighboring pixels.
The filter accepts the following options:
Set the filter bitmap file, which can be any image format supported by libavformat. The width and height of the image file must match those of the video stream being processed.
Pixels in the provided bitmap image with a value of zero are not considered part of the logo, non-zero pixels are considered part of the logo. If you use white (255) for the logo and black (0) for the rest, you will be safe. For making the filter bitmap, it is recommended to take a screen capture of a black frame with the logo visible, and then using a threshold filter followed by the erode filter once or twice.
If needed, little splotches can be fixed manually. Remember that if logo pixels are not covered, the filter quality will be much reduced. Marking too many pixels as part of the logo does not hurt as much, but it will increase the amount of blurring needed to cover over the image and will destroy more information than necessary, and extra pixels will slow things down on a large logo.
This filter uses the repeat_field flag from the Video ES headers and hard repeats fields based on its value.
Reverse a video clip.
Warning: This filter requires memory to buffer the entire clip, so trimming is suggested.
trim=end=5,reverse
Shift R/G/B/A pixels horizontally and/or vertically.
The filter accepts the following options:
Set amount to shift red horizontally.
Set amount to shift red vertically.
Set amount to shift green horizontally.
Set amount to shift green vertically.
Set amount to shift blue horizontally.
Set amount to shift blue vertically.
Set amount to shift alpha horizontally.
Set amount to shift alpha vertically.
Set edge mode, can be smear, default, or warp.
This filter supports the all above options as commands.
Apply roberts cross operator to input video stream.
The filter accepts the following option:
Set which planes will be processed, unprocessed planes will be copied. By default value 0xf, all planes will be processed.
Set value which will be multiplied with filtered result.
Set value which will be added to filtered result.
This filter supports the all above options as commands.
Rotate video by an arbitrary angle expressed in radians.
The filter accepts the following options:
A description of the optional parameters follows.
Set an expression for the angle by which to rotate the input video clockwise, expressed as a number of radians. A negative value will result in a counter-clockwise rotation. By default it is set to "0".
This expression is evaluated for each frame.
Set the output width expression, default value is "iw". This expression is evaluated just once during configuration.
Set the output height expression, default value is "ih". This expression is evaluated just once during configuration.
Enable bilinear interpolation if set to 1, a value of 0 disables it. Default value is 1.
Set the color used to fill the output area not covered by the rotated image. For the general syntax of this option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual. If the special value "none" is selected then no background is printed (useful for example if the background is never shown).
Default value is "black".
The expressions for the angle and the output size can contain the following constants and functions:
sequential number of the input frame, starting from 0. It is always NAN before the first frame is filtered.
time in seconds of the input frame, it is set to 0 when the filter is configured. It is always NAN before the first frame is filtered.
horizontal and vertical chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
the input video width and height
the output width and height, that is the size of the padded area as specified by the width and height expressions
the minimal width/height required for completely containing the input video rotated by a radians.
These are only available when computing the out_w and out_h expressions.
rotate=PI/6
rotate=-PI/6
rotate=45*PI/180
rotate=PI/3+2*PI*t/T
rotate=A*sin(2*PI/T*t)
rotate='2*PI*t:ow=hypot(iw,ih):oh=ow'
rotate=2*PI*t:ow='min(iw,ih)/sqrt(2)':oh=ow:c=none
The filter supports the following commands:
Set the angle expression. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Apply Shape Adaptive Blur.
The filter accepts the following options:
Set luma blur filter strength, must be a value in range 0.1-4.0, default value is 1.0. A greater value will result in a more blurred image, and in slower processing.
Set luma pre-filter radius, must be a value in the 0.1-2.0 range, default value is 1.0.
Set luma maximum difference between pixels to still be considered, must be a value in the 0.1-100.0 range, default value is 1.0.
Set chroma blur filter strength, must be a value in range -0.9-4.0. A greater value will result in a more blurred image, and in slower processing.
Set chroma pre-filter radius, must be a value in the -0.9-2.0 range.
Set chroma maximum difference between pixels to still be considered, must be a value in the -0.9-100.0 range.
Each chroma option value, if not explicitly specified, is set to the corresponding luma option value.
Scale (resize) the input video, using the libswscale library.
The scale filter forces the output display aspect ratio to be the same of the input, by changing the output sample aspect ratio.
If the input image format is different from the format requested by the next filter, the scale filter will convert the input to the requested format.
The filter accepts the following options, or any of the options supported by the libswscale scaler.
See (ffmpeg-scaler)the ffmpeg-scaler manual for the complete list of scaler options.
Set the output video dimension expression. Default value is the input dimension.
If the width or w value is 0, the input width is used for the output. If the height or h value is 0, the input height is used for the output.
If one and only one of the values is -n with n >= 1, the scale filter will use a value that maintains the aspect ratio of the input image, calculated from the other specified dimension. After that it will, however, make sure that the calculated dimension is divisible by n and adjust the value if necessary.
If both values are -n with n >= 1, the behavior will be identical to both values being set to 0 as previously detailed.
See below for the list of accepted constants for use in the dimension expression.
Specify when to evaluate width and height expression. It accepts the following values:
Only evaluate expressions once during the filter initialization or when a command is processed.
Evaluate expressions for each incoming frame.
Default value is ‘init’.
Set the interlacing mode. It accepts the following values:
Force interlaced aware scaling.
Do not apply interlaced scaling.
Select interlaced aware scaling depending on whether the source frames are flagged as interlaced or not.
Default value is ‘0’.
Set libswscale scaling flags. See (ffmpeg-scaler)the ffmpeg-scaler manual for the complete list of values. If not explicitly specified the filter applies the default flags.
Set libswscale input parameters for scaling algorithms that need them. See (ffmpeg-scaler)the ffmpeg-scaler manual for the complete documentation. If not explicitly specified the filter applies empty parameters.
Set the video size. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Set in/output YCbCr color space type.
This allows the autodetected value to be overridden as well as allows forcing a specific value used for the output and encoder.
If not specified, the color space type depends on the pixel format.
Possible values:
Choose automatically.
Format conforming to International Telecommunication Union (ITU) Recommendation BT.709.
Set color space conforming to the United States Federal Communications Commission (FCC) Code of Federal Regulations (CFR) Title 47 (2003) 73.682 (a).
Set color space conforming to:
Set color space conforming to SMPTE ST 240:1999.
Set color space conforming to ITU-R BT.2020 non-constant luminance system.
Set in/output YCbCr sample range.
This allows the autodetected value to be overridden as well as allows forcing a specific value used for the output and encoder. If not specified, the range depends on the pixel format. Possible values:
Choose automatically.
Set full range (0-255 in case of 8-bit luma).
Set "MPEG" range (16-235 in case of 8-bit luma).
Enable decreasing or increasing output video width or height if necessary to keep the original aspect ratio. Possible values:
Scale the video as specified and disable this feature.
The output video dimensions will automatically be decreased if needed.
The output video dimensions will automatically be increased if needed.
One useful instance of this option is that when you know a specific device’s maximum allowed resolution, you can use this to limit the output video to that, while retaining the aspect ratio. For example, device A allows 1280x720 playback, and your video is 1920x800. Using this option (set it to decrease) and specifying 1280x720 to the command line makes the output 1280x533.
Please note that this is a different thing than specifying -1 for w or h, you still need to specify the output resolution for this option to work.
Ensures that both the output dimensions, width and height, are divisible by the
given integer when used together with force_original_aspect_ratio. This
works similar to using -n
in the w and h options.
This option respects the value set for force_original_aspect_ratio, increasing or decreasing the resolution accordingly. The video’s aspect ratio may be slightly modified.
This option can be handy if you need to have a video fit within or exceed a defined resolution using force_original_aspect_ratio but also have encoder restrictions on width or height divisibility.
The values of the w and h options are expressions containing the following constants:
The input width and height
These are the same as in_w and in_h.
The output (scaled) width and height
These are the same as out_w and out_h
The same as iw / ih
input sample aspect ratio
The input display aspect ratio. Calculated from (iw / ih) * sar
.
horizontal and vertical input chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
horizontal and vertical output chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
The (sequential) number of the input frame, starting from 0.
Only available with eval=frame
.
The presentation timestamp of the input frame, expressed as a number of
seconds. Only available with eval=frame
.
The position (byte offset) of the frame in the input stream, or NaN if
this information is unavailable and/or meaningless (for example in case of synthetic video).
Only available with eval=frame
.
Deprecated, do not use.
scale=w=200:h=100
This is equivalent to:
scale=200:100
or:
scale=200x100
scale=qcif
which can also be written as:
scale=size=qcif
scale=w=2*iw:h=2*ih
scale=2*in_w:2*in_h
scale=2*iw:2*ih:interl=1
scale=w=iw/2:h=ih/2
scale=3/2*iw:ow
scale=iw:1/PHI*iw scale=ih*PHI:ih
scale=w=3/2*oh:h=3/5*ih
scale="trunc(3/2*iw/hsub)*hsub:trunc(3/2*ih/vsub)*vsub"
scale=w='min(500\, iw*3/2):h=-1'
scale='trunc(ih*dar):ih',setsar=1/1
scale='trunc(ih*dar/2)*2:trunc(ih/2)*2',setsar=1/1
Scale (resize) and convert (pixel format) the input video, using accelerated CUDA kernels. Setting the output width and height works in the same way as for the scale filter.
The filter accepts the following options:
Set the output video dimension expression. Default value is the input dimension.
Allows for the same expressions as the scale filter.
Sets the algorithm used for scaling:
Nearest neighbour
Used by default if input parameters match the desired output.
Bilinear
Bicubic
This is the default.
Lanczos
Controls the output pixel format. By default, or if none is specified, the input pixel format is used.
The filter does not support converting between YUV and RGB pixel formats.
If set to 0, every frame is processed, even if no conversion is necessary. This mode can be useful to use the filter as a buffer for a downstream frame-consumer that exhausts the limited decoder frame pool.
If set to 1, frames are passed through as-is if they match the desired output parameters. This is the default behaviour.
Algorithm-Specific parameter.
Affects the curves of the bicubic algorithm.
Work the same as the identical scale filter options.
scale_cuda=-2:720:format=yuv420p
scale_cuda=4096:2160:interp_algo=nearest
scale_cuda=passthrough=0
Use the NVIDIA Performance Primitives (libnpp) to perform scaling and/or pixel format conversion on CUDA video frames. Setting the output width and height works in the same way as for the scale filter.
The following additional options are accepted:
The pixel format of the output CUDA frames. If set to the string "same" (the default), the input format will be kept. Note that automatic format negotiation and conversion is not yet supported for hardware frames
The interpolation algorithm used for resizing. One of the following:
Nearest neighbour.
2-parameter cubic (B=1, C=0)
2-parameter cubic (B=0, C=1/2)
2-parameter cubic (B=1/2, C=3/10)
Supersampling
Enable decreasing or increasing output video width or height if necessary to keep the original aspect ratio. Possible values:
Scale the video as specified and disable this feature.
The output video dimensions will automatically be decreased if needed.
The output video dimensions will automatically be increased if needed.
One useful instance of this option is that when you know a specific device’s maximum allowed resolution, you can use this to limit the output video to that, while retaining the aspect ratio. For example, device A allows 1280x720 playback, and your video is 1920x800. Using this option (set it to decrease) and specifying 1280x720 to the command line makes the output 1280x533.
Please note that this is a different thing than specifying -1 for w or h, you still need to specify the output resolution for this option to work.
Ensures that both the output dimensions, width and height, are divisible by the
given integer when used together with force_original_aspect_ratio. This
works similar to using -n
in the w and h options.
This option respects the value set for force_original_aspect_ratio, increasing or decreasing the resolution accordingly. The video’s aspect ratio may be slightly modified.
This option can be handy if you need to have a video fit within or exceed a defined resolution using force_original_aspect_ratio but also have encoder restrictions on width or height divisibility.
Specify when to evaluate width and height expression. It accepts the following values:
Only evaluate expressions once during the filter initialization or when a command is processed.
Evaluate expressions for each incoming frame.
The values of the w and h options are expressions containing the following constants:
The input width and height
These are the same as in_w and in_h.
The output (scaled) width and height
These are the same as out_w and out_h
The same as iw / ih
input sample aspect ratio
The input display aspect ratio. Calculated from (iw / ih) * sar
.
The (sequential) number of the input frame, starting from 0.
Only available with eval=frame
.
The presentation timestamp of the input frame, expressed as a number of
seconds. Only available with eval=frame
.
The position (byte offset) of the frame in the input stream, or NaN if
this information is unavailable and/or meaningless (for example in case of synthetic video).
Only available with eval=frame
.
Deprecated, do not use.
Scale (resize) the input video, based on a reference video.
See the scale filter for available options, scale2ref supports the same but uses the reference video instead of the main input as basis. scale2ref also supports the following additional constants for the w and h options:
The main input video’s width and height
The same as main_w / main_h
The main input video’s sample aspect ratio
The main input video’s display aspect ratio. Calculated from
(main_w / main_h) * main_sar
.
The main input video’s horizontal and vertical chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
The (sequential) number of the main input frame, starting from 0.
Only available with eval=frame
.
The presentation timestamp of the main input frame, expressed as a number of
seconds. Only available with eval=frame
.
The position (byte offset) of the frame in the main input stream, or NaN if
this information is unavailable and/or meaningless (for example in case of synthetic video).
Only available with eval=frame
.
'scale2ref[b][a];[a][b]overlay'
[logo-in][video-in]scale2ref=w=oh*mdar:h=ih/10[logo-out][video-out]
This filter supports the following commands:
Set the output video dimension expression. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Use the NVIDIA Performance Primitives (libnpp) to scale (resize) the input video, based on a reference video.
See the scale_npp filter for available options, scale2ref_npp supports the same but uses the reference video instead of the main input as basis. scale2ref_npp also supports the following additional constants for the w and h options:
The main input video’s width and height
The same as main_w / main_h
The main input video’s sample aspect ratio
The main input video’s display aspect ratio. Calculated from
(main_w / main_h) * main_sar
.
The (sequential) number of the main input frame, starting from 0.
Only available with eval=frame
.
The presentation timestamp of the main input frame, expressed as a number of
seconds. Only available with eval=frame
.
The position (byte offset) of the frame in the main input stream, or NaN if
this information is unavailable and/or meaningless (for example in case of synthetic video).
Only available with eval=frame
.
'scale2ref_npp[b][a];[a][b]overlay_cuda'
[logo-in][video-in]scale2ref_npp=w=oh*mdar:h=ih/10[logo-out][video-out]
Scale and convert the color parameters using VTPixelTransferSession.
The filter accepts the following options:
Set the output video dimension expression. Default value is the input dimension.
Set the output colorspace matrix.
Set the output color primaries.
Set the output transfer characteristics.
Apply scharr operator to input video stream.
The filter accepts the following option:
Set which planes will be processed, unprocessed planes will be copied. By default value 0xf, all planes will be processed.
Set value which will be multiplied with filtered result.
Set value which will be added to filtered result.
This filter supports the all above options as commands.
Scroll input video horizontally and/or vertically by constant speed.
The filter accepts the following options:
Set the horizontal scrolling speed. Default is 0. Allowed range is from -1 to 1. Negative values changes scrolling direction.
Set the vertical scrolling speed. Default is 0. Allowed range is from -1 to 1. Negative values changes scrolling direction.
Set the initial horizontal scrolling position. Default is 0. Allowed range is from 0 to 1.
Set the initial vertical scrolling position. Default is 0. Allowed range is from 0 to 1.
This filter supports the following commands:
Set the horizontal scrolling speed.
Set the vertical scrolling speed.
Detect video scene change.
This filter sets frame metadata with mafd between frame, the scene score, and forward the frame to the next filter, so they can use these metadata to detect scene change or others.
In addition, this filter logs a message and sets frame metadata when it detects a scene change by threshold.
lavfi.scd.mafd
metadata keys are set with mafd for every frame.
lavfi.scd.score
metadata keys are set with scene change score for every frame
to detect scene change.
lavfi.scd.time
metadata keys are set with current filtered frame time which
detect scene change with threshold.
The filter accepts the following options:
Set the scene change detection threshold as a percentage of maximum change. Good
values are in the [8.0, 14.0]
range. The range for threshold is
[0., 100.]
.
Default value is 10.
.
Set the flag to pass scene change frames to the next filter. Default value is 0
You can enable it if you want to get snapshot of scene change frames only.
Adjust cyan, magenta, yellow and black (CMYK) to certain ranges of colors (such as "reds", "yellows", "greens", "cyans", ...). The adjustment range is defined by the "purity" of the color (that is, how saturated it already is).
This filter is similar to the Adobe Photoshop Selective Color tool.
The filter accepts the following options:
Select color correction method.
Available values are:
Specified adjustments are applied "as-is" (added/subtracted to original pixel component value).
Specified adjustments are relative to the original component value.
Default is absolute
.
Adjustments for red pixels (pixels where the red component is the maximum)
Adjustments for yellow pixels (pixels where the blue component is the minimum)
Adjustments for green pixels (pixels where the green component is the maximum)
Adjustments for cyan pixels (pixels where the red component is the minimum)
Adjustments for blue pixels (pixels where the blue component is the maximum)
Adjustments for magenta pixels (pixels where the green component is the minimum)
Adjustments for white pixels (pixels where all components are greater than 128)
Adjustments for all pixels except pure black and pure white
Adjustments for black pixels (pixels where all components are lesser than 128)
Specify a Photoshop selective color file (.asv
) to import the settings from.
All the adjustment settings (reds, yellows, ...) accept up to 4 space separated floating point adjustment values in the [-1,1] range, respectively to adjust the amount of cyan, magenta, yellow and black for the pixels of its range.
The separatefields
takes a frame-based video input and splits
each frame into its components fields, producing a new half height clip
with twice the frame rate and twice the frame count.
This filter use field-dominance information in frame to decide which
of each pair of fields to place first in the output.
If it gets it wrong use setfield filter before separatefields
filter.
The setdar
filter sets the Display Aspect Ratio for the filter
output video.
This is done by changing the specified Sample (aka Pixel) Aspect Ratio, according to the following equation:
DAR = HORIZONTAL_RESOLUTION / VERTICAL_RESOLUTION * SAR
Keep in mind that the setdar
filter does not modify the pixel
dimensions of the video frame. Also, the display aspect ratio set by
this filter may be changed by later filters in the filterchain,
e.g. in case of scaling or if another "setdar" or a "setsar" filter is
applied.
The setsar
filter sets the Sample (aka Pixel) Aspect Ratio for
the filter output video.
Note that as a consequence of the application of this filter, the output display aspect ratio will change according to the equation above.
Keep in mind that the sample aspect ratio set by the setsar
filter may be changed by later filters in the filterchain, e.g. if
another "setsar" or a "setdar" filter is applied.
It accepts the following parameters:
setdar
only), sar (setsar
only)Set the aspect ratio used by the filter.
The parameter can be a floating point number string, or an expression. If the parameter is not specified, the value "0" is assumed, meaning that the same input value is used.
Set the maximum integer value to use for expressing numerator and
denominator when reducing the expressed aspect ratio to a rational.
Default value is 100
.
The parameter sar is an expression containing the following constants:
The input width and height.
Same as w / h.
The input sample aspect ratio.
The input display aspect ratio. It is the same as (w / h) * sar.
Horizontal and vertical chroma subsample values. For example, for the pixel format "yuv422p" hsub is 2 and vsub is 1.
setdar=dar=1.77777 setdar=dar=16/9
setsar=sar=10/11
setdar=ratio=16/9:max=1000
Force field for the output video frame.
The setfield
filter marks the interlace type field for the
output frames. It does not change the input frame, but only sets the
corresponding property, which affects how the frame is treated by
following filters (e.g. fieldorder
or yadif
).
The filter accepts the following options:
Available values are:
Keep the same field property.
Mark the frame as bottom-field-first.
Mark the frame as top-field-first.
Mark the frame as progressive.
Force frame parameter for the output video frame.
The setparams
filter marks interlace and color range for the
output frames. It does not change the input frame, but only sets the
corresponding property, which affects how the frame is treated by
filters/encoders.
Available values are:
Keep the same field property (default).
Mark the frame as bottom-field-first.
Mark the frame as top-field-first.
Mark the frame as progressive.
Available values are:
Keep the same color range property (default).
Mark the frame as unspecified color range.
Mark the frame as limited range.
Mark the frame as full range.
Set the color primaries. Available values are:
Keep the same color primaries property (default).
Set the color transfer. Available values are:
Keep the same color trc property (default).
Set the colorspace. Available values are:
Keep the same colorspace property (default).
Use the NVIDIA Performance Primitives (libnpp) to perform image sharpening with border control.
The following additional options are accepted:
Type of sampling to be used ad frame borders. One of the following:
Replicate pixel values.
Apply shear transform to input video.
This filter supports the following options:
Shear factor in X-direction. Default value is 0. Allowed range is from -2 to 2.
Shear factor in Y-direction. Default value is 0. Allowed range is from -2 to 2.
Set the color used to fill the output area not covered by the transformed video. For the general syntax of this option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual. If the special value "none" is selected then no background is printed (useful for example if the background is never shown).
Default value is "black".
Set interpolation type. Can be bilinear
or nearest
. Default is bilinear
.
This filter supports the all above options as commands.
Show a line containing various information for each input video frame. The input video is not modified.
This filter supports the following options:
Calculate checksums of each plane. By default enabled.
The shown line contains a sequence of key/value pairs of the form key:value.
The following values are shown in the output:
The (sequential) number of the input frame, starting from 0.
The Presentation TimeStamp of the input frame, expressed as a number of time base units. The time base unit depends on the filter input pad.
The Presentation TimeStamp of the input frame, expressed as a number of seconds.
The pixel format name.
The sample aspect ratio of the input frame, expressed in the form num/den.
The size of the input frame. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
The type of interlaced mode ("P" for "progressive", "T" for top field first, "B" for bottom field first).
This is 1 if the frame is a key frame, 0 otherwise.
The picture type of the input frame ("I" for an I-frame, "P" for a
P-frame, "B" for a B-frame, or "?" for an unknown type).
Also refer to the documentation of the AVPictureType
enum and of
the av_get_picture_type_char
function defined in
libavutil/avutil.h.
The Adler-32 checksum (printed in hexadecimal) of all the planes of the input frame.
The Adler-32 checksum (printed in hexadecimal) of each plane of the input frame, expressed in the form "[c0 c1 c2 c3]".
The mean value of pixels in each plane of the input frame, expressed in the form "[mean0 mean1 mean2 mean3]".
The standard deviation of pixel values in each plane of the input frame, expressed in the form "[stdev0 stdev1 stdev2 stdev3]".
Displays the 256 colors palette of each frame. This filter is only relevant for pal8 pixel format frames.
It accepts the following option:
Set the size of the box used to represent one palette color entry. Default is
30
(for a 30x30
pixel box).
Reorder and/or duplicate and/or drop video frames.
It accepts the following parameters:
Set the destination indexes of input frames. This is space or ’|’ separated list of indexes that maps input frames to output frames. Number of indexes also sets maximal value that each index may have. ’-1’ index have special meaning and that is to drop frame.
The first frame has the index 0. The default is to keep the input unchanged.
ffmpeg -i INPUT -vf "shuffleframes=0 2 1" OUTPUT
ffmpeg -i INPUT -vf "shuffleframes=9 1 2 3 4 5 6 7 8 0" OUTPUT
Reorder pixels in video frames.
This filter accepts the following options:
Set shuffle direction. Can be forward or inverse direction. Default direction is forward.
Set shuffle mode. Can be horizontal, vertical or block mode.
Set shuffle block_size. In case of horizontal shuffle mode only width part of size is used, and in case of vertical shuffle mode only height part of size is used.
Set random seed used with shuffling pixels. Mainly useful to set to be able to reverse filtering process to get original input. For example, to reverse forward shuffle you need to use same parameters and exact same seed and to set direction to inverse.
Reorder and/or duplicate video planes.
It accepts the following parameters:
The index of the input plane to be used as the first output plane.
The index of the input plane to be used as the second output plane.
The index of the input plane to be used as the third output plane.
The index of the input plane to be used as the fourth output plane.
The first plane has the index 0. The default is to keep the input unchanged.
Evaluate various visual metrics that assist in determining issues associated with the digitization of analog video media.
By default the filter will log these metadata values:
Display the minimal Y value contained within the input frame. Expressed in range of [0-255].
Display the Y value at the 10% percentile within the input frame. Expressed in range of [0-255].
Display the average Y value within the input frame. Expressed in range of [0-255].
Display the Y value at the 90% percentile within the input frame. Expressed in range of [0-255].
Display the maximum Y value contained within the input frame. Expressed in range of [0-255].
Display the minimal U value contained within the input frame. Expressed in range of [0-255].
Display the U value at the 10% percentile within the input frame. Expressed in range of [0-255].
Display the average U value within the input frame. Expressed in range of [0-255].
Display the U value at the 90% percentile within the input frame. Expressed in range of [0-255].
Display the maximum U value contained within the input frame. Expressed in range of [0-255].
Display the minimal V value contained within the input frame. Expressed in range of [0-255].
Display the V value at the 10% percentile within the input frame. Expressed in range of [0-255].
Display the average V value within the input frame. Expressed in range of [0-255].
Display the V value at the 90% percentile within the input frame. Expressed in range of [0-255].
Display the maximum V value contained within the input frame. Expressed in range of [0-255].
Display the minimal saturation value contained within the input frame. Expressed in range of [0-~181.02].
Display the saturation value at the 10% percentile within the input frame. Expressed in range of [0-~181.02].
Display the average saturation value within the input frame. Expressed in range of [0-~181.02].
Display the saturation value at the 90% percentile within the input frame. Expressed in range of [0-~181.02].
Display the maximum saturation value contained within the input frame. Expressed in range of [0-~181.02].
Display the median value for hue within the input frame. Expressed in range of [0-360].
Display the average value for hue within the input frame. Expressed in range of [0-360].
Display the average of sample value difference between all values of the Y plane in the current frame and corresponding values of the previous input frame. Expressed in range of [0-255].
Display the average of sample value difference between all values of the U plane in the current frame and corresponding values of the previous input frame. Expressed in range of [0-255].
Display the average of sample value difference between all values of the V plane in the current frame and corresponding values of the previous input frame. Expressed in range of [0-255].
Display bit depth of Y plane in current frame. Expressed in range of [0-16].
Display bit depth of U plane in current frame. Expressed in range of [0-16].
Display bit depth of V plane in current frame. Expressed in range of [0-16].
The filter accepts the following options:
stat specify an additional form of image analysis. out output video with the specified type of pixel highlighted.
Both options accept the following values:
Identify temporal outliers pixels. A temporal outlier is a pixel unlike the neighboring pixels of the same field. Examples of temporal outliers include the results of video dropouts, head clogs, or tape tracking issues.
Identify vertical line repetition. Vertical line repetition includes similar rows of pixels within a frame. In born-digital video vertical line repetition is common, but this pattern is uncommon in video digitized from an analog source. When it occurs in video that results from the digitization of an analog source it can indicate concealment from a dropout compensator.
Identify pixels that fall outside of legal broadcast range.
Set the highlight color for the out option. The default color is yellow.
ffprobe -f lavfi movie=example.mov,signalstats="stat=tout+vrep+brng" -show_frames
ffprobe -f lavfi movie=example.mov,signalstats -show_entries frame_tags=lavfi.signalstats.YMAX,lavfi.signalstats.YMIN
ffplay example.mov -vf signalstats="out=brng:color=red"
ffplay example.mov -vf signalstats=stat=brng+vrep+tout,drawtext=fontfile=FreeSerif.ttf:textfile=signalstat_drawtext.txt
The contents of signalstat_drawtext.txt used in the command are:
time %{pts:hms} Y (%{metadata:lavfi.signalstats.YMIN}-%{metadata:lavfi.signalstats.YMAX}) U (%{metadata:lavfi.signalstats.UMIN}-%{metadata:lavfi.signalstats.UMAX}) V (%{metadata:lavfi.signalstats.VMIN}-%{metadata:lavfi.signalstats.VMAX}) saturation maximum: %{metadata:lavfi.signalstats.SATMAX}
Calculates the MPEG-7 Video Signature. The filter can handle more than one input. In this case the matching between the inputs can be calculated additionally. The filter always passes through the first input. The signature of each stream can be written into a file.
It accepts the following options:
Enable or disable the matching process.
Available values are:
Disable the calculation of a matching (default).
Calculate the matching for the whole video and output whether the whole video matches or only parts.
Calculate only until a matching is found or the video ends. Should be faster in some cases.
Set the number of inputs. The option value must be a non negative integer. Default value is 1.
Set the path to which the output is written. If there is more than one input, the path must be a prototype, i.e. must contain %d or %0nd (where n is a positive integer), that will be replaced with the input number. If no filename is specified, no output will be written. This is the default.
Choose the output format.
Available values are:
Use the specified binary representation (default).
Use the specified xml representation.
Set threshold to detect one word as similar. The option value must be an integer greater than zero. The default value is 9000.
Set threshold to detect all words as similar. The option value must be an integer greater than zero. The default value is 60000.
Set threshold to detect frames as similar. The option value must be an integer greater than zero. The default value is 116.
Set the minimum length of a sequence in frames to recognize it as matching sequence. The option value must be a non negative integer value. The default value is 0.
Set the minimum relation, that matching frames to all frames must have. The option value must be a double value between 0 and 1. The default value is 0.5.
ffmpeg -i input.mkv -vf signature=filename=signature.bin -map 0:v -f null -
ffmpeg -i input1.mkv -i input2.mkv -filter_complex "[0:v][1:v] signature=nb_inputs=2:detectmode=full:format=xml:filename=signature%d.xml" -map :v -f null -
Calculate Spatial Information (SI) and Temporal Information (TI) scores for a video, as defined in ITU-T Rec. P.910 (11/21): Subjective video quality assessment methods for multimedia applications. Available PDF at https://www.itu.int/rec/T-REC-P.910-202111-S/en. Note that this is a legacy implementation that corresponds to a superseded recommendation. Refer to ITU-T Rec. P.910 (07/22) for the latest version: https://www.itu.int/rec/T-REC-P.910-202207-I/en
It accepts the following option:
If set to 1, Summary statistics will be printed to the console. Default 0.
Blur the input video without impacting the outlines.
It accepts the following options:
Set the luma radius. The option value must be a float number in the range [0.1,5.0] that specifies the variance of the gaussian filter used to blur the image (slower if larger). Default value is 1.0.
Set the luma strength. The option value must be a float number in the range [-1.0,1.0] that configures the blurring. A value included in [0.0,1.0] will blur the image whereas a value included in [-1.0,0.0] will sharpen the image. Default value is 1.0.
Set the luma threshold used as a coefficient to determine whether a pixel should be blurred or not. The option value must be an integer in the range [-30,30]. A value of 0 will filter all the image, a value included in [0,30] will filter flat areas and a value included in [-30,0] will filter edges. Default value is 0.
Set the chroma radius. The option value must be a float number in the range [0.1,5.0] that specifies the variance of the gaussian filter used to blur the image (slower if larger). Default value is luma_radius.
Set the chroma strength. The option value must be a float number in the range [-1.0,1.0] that configures the blurring. A value included in [0.0,1.0] will blur the image whereas a value included in [-1.0,0.0] will sharpen the image. Default value is luma_strength.
Set the chroma threshold used as a coefficient to determine whether a pixel should be blurred or not. The option value must be an integer in the range [-30,30]. A value of 0 will filter all the image, a value included in [0,30] will filter flat areas and a value included in [-30,0] will filter edges. Default value is luma_threshold.
If a chroma option is not explicitly set, the corresponding luma value is set.
Apply sobel operator to input video stream.
The filter accepts the following option:
Set which planes will be processed, unprocessed planes will be copied. By default value 0xf, all planes will be processed.
Set value which will be multiplied with filtered result.
Set value which will be added to filtered result.
Apply a simple postprocessing filter that compresses and decompresses the image
at several (or - in the case of quality level 6
- all) shifts
and average the results.
The filter accepts the following options:
Set quality. This option defines the number of levels for averaging. It accepts
an integer in the range 0-6. If set to 0
, the filter will have no
effect. A value of 6
means the higher quality. For each increment of
that value the speed drops by a factor of approximately 2. Default value is
3
.
Force a constant quantization parameter. If not set, the filter will use the QP from the video stream (if available).
Set thresholding mode. Available modes are:
Set hard thresholding (default).
Set soft thresholding (better de-ringing effect, but likely blurrier).
Enable the use of the QP from the B-Frames if set to 1
. Using this
option may cause flicker since the B-Frames have often larger QP. Default is
0
(not enabled).
Scale the input by applying one of the super-resolution methods based on convolutional neural networks. Supported models:
Training scripts as well as scripts for model file (.pb) saving can be found at https://github.com/XueweiMeng/sr/tree/sr_dnn_native. Original repository is at https://github.com/HighVoltageRocknRoll/sr.git.
The filter accepts the following options:
Specify which DNN backend to use for model loading and execution. This option accepts the following values:
TensorFlow backend. To enable this backend you
need to install the TensorFlow for C library (see
https://www.tensorflow.org/install/lang_c) and configure FFmpeg with
--enable-libtensorflow
Set path to model file specifying network architecture and its parameters. Note that different backends use different file formats. TensorFlow, OpenVINO backend can load files for only its format.
Set scale factor for SRCNN model. Allowed values are 2
, 3
and 4
.
Default value is 2
. Scale factor is necessary for SRCNN model, because it accepts
input upscaled using bicubic upscaling with proper scale factor.
To get full functionality (such as async execution), please use the dnn_processing filter.
Obtain the SSIM (Structural SImilarity Metric) between two input videos.
This filter takes in input two input videos, the first input is considered the "main" source and is passed unchanged to the output. The second input is used as a "reference" video for computing the SSIM.
Both video inputs must have the same resolution and pixel format for this filter to work correctly. Also it assumes that both inputs have the same number of frames, which are compared one by one.
The filter stores the calculated SSIM of each frame.
The description of the accepted parameters follows.
If specified the filter will use the named file to save the SSIM of each individual frame. When filename equals "-" the data is sent to standard output.
The file printed if stats_file is selected, contains a sequence of key/value pairs of the form key:value for each compared couple of frames.
A description of each shown parameter follows:
sequential number of the input frame, starting from 1
SSIM of the compared frames for the component specified by the suffix.
SSIM of the compared frames for the whole frame.
Same as above but in dB representation.
This filter also supports the framesync options.
movie=ref_movie.mpg, setpts=PTS-STARTPTS [main]; [main][ref] ssim="stats_file=stats.log" [out]
On this example the input file being processed is compared with the reference file ref_movie.mpg. The SSIM of each individual frame is stored in stats.log.
ffmpeg -i main.mpg -i ref.mpg -lavfi "ssim;[0:v][1:v]psnr" -f null -
ffmpeg -i main.mpg -i ref.mkv -lavfi "[0:v]settb=AVTB,setpts=PTS-STARTPTS[main];[1:v]settb=AVTB,setpts=PTS-STARTPTS[ref];[main][ref]ssim" -f null -
Convert between different stereoscopic image formats.
The filters accept the following options:
Set stereoscopic image format of input.
Available values for input image formats are:
side by side parallel (left eye left, right eye right)
side by side crosseye (right eye left, left eye right)
side by side parallel with half width resolution (left eye left, right eye right)
side by side crosseye with half width resolution (right eye left, left eye right)
above-below (left eye above, right eye below)
above-below (right eye above, left eye below)
above-below with half height resolution (left eye above, right eye below)
above-below with half height resolution (right eye above, left eye below)
alternating frames (left eye first, right eye second)
alternating frames (right eye first, left eye second)
interleaved rows (left eye has top row, right eye starts on next row)
interleaved rows (right eye has top row, left eye starts on next row)
interleaved columns, left eye first
interleaved columns, right eye first
Default value is ‘sbsl’.
Set stereoscopic image format of output.
side by side parallel (left eye left, right eye right)
side by side crosseye (right eye left, left eye right)
side by side parallel with half width resolution (left eye left, right eye right)
side by side crosseye with half width resolution (right eye left, left eye right)
above-below (left eye above, right eye below)
above-below (right eye above, left eye below)
above-below with half height resolution (left eye above, right eye below)
above-below with half height resolution (right eye above, left eye below)
alternating frames (left eye first, right eye second)
alternating frames (right eye first, left eye second)
interleaved rows (left eye has top row, right eye starts on next row)
interleaved rows (right eye has top row, left eye starts on next row)
anaglyph red/blue gray (red filter on left eye, blue filter on right eye)
anaglyph red/green gray (red filter on left eye, green filter on right eye)
anaglyph red/cyan gray (red filter on left eye, cyan filter on right eye)
anaglyph red/cyan half colored (red filter on left eye, cyan filter on right eye)
anaglyph red/cyan color (red filter on left eye, cyan filter on right eye)
anaglyph red/cyan color optimized with the least squares projection of dubois (red filter on left eye, cyan filter on right eye)
anaglyph green/magenta gray (green filter on left eye, magenta filter on right eye)
anaglyph green/magenta half colored (green filter on left eye, magenta filter on right eye)
anaglyph green/magenta colored (green filter on left eye, magenta filter on right eye)
anaglyph green/magenta color optimized with the least squares projection of dubois (green filter on left eye, magenta filter on right eye)
anaglyph yellow/blue gray (yellow filter on left eye, blue filter on right eye)
anaglyph yellow/blue half colored (yellow filter on left eye, blue filter on right eye)
anaglyph yellow/blue colored (yellow filter on left eye, blue filter on right eye)
anaglyph yellow/blue color optimized with the least squares projection of dubois (yellow filter on left eye, blue filter on right eye)
mono output (left eye only)
mono output (right eye only)
checkerboard, left eye first
checkerboard, right eye first
interleaved columns, left eye first
interleaved columns, right eye first
HDMI frame pack
Default value is ‘arcd’.
stereo3d=sbsl:aybd
stereo3d=abl:sbsr
Select video or audio streams.
The filter accepts the following options:
Set number of inputs. Default is 2.
Set input indexes to remap to outputs.
The streamselect
and astreamselect
filter supports the following
commands:
Set input indexes to remap to outputs.
Draw subtitles on top of input video using the libass library.
To enable compilation of this filter you need to configure FFmpeg with
--enable-libass
. This filter also requires a build with libavcodec and
libavformat to convert the passed subtitles file to ASS (Advanced Substation
Alpha) subtitles format.
The filter accepts the following options:
Set the filename of the subtitle file to read. It must be specified.
Specify the size of the original video, the video for which the ASS file was composed. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual. Due to a misdesign in ASS aspect ratio arithmetic, this is necessary to correctly scale the fonts if the aspect ratio has been changed.
Set a directory path containing fonts that can be used by the filter. These fonts will be used in addition to whatever the font provider uses.
Process alpha channel, by default alpha channel is untouched.
Set subtitles input character encoding. subtitles
filter only. Only
useful if not UTF-8.
Set subtitles stream index. subtitles
filter only.
Override default style or script info parameters of the subtitles. It accepts a
string containing ASS style format KEY=VALUE
couples separated by ",".
Break lines according to the Unicode Line Breaking Algorithm. Availability requires at least libass release 0.17.0 (or LIBASS_VERSION 0x01600010), and libass must have been built with libunibreak.
The option is enabled by default except for native ASS.
If the first key is not specified, it is assumed that the first value specifies the filename.
For example, to render the file sub.srt on top of the input video, use the command:
subtitles=sub.srt
which is equivalent to:
subtitles=filename=sub.srt
To render the default subtitles stream from file video.mkv, use:
subtitles=video.mkv
To render the second subtitles stream from that file, use:
subtitles=video.mkv:si=1
To make the subtitles stream from sub.srt appear in 80% transparent blue
DejaVu Serif
, use:
subtitles=sub.srt:force_style='Fontname=DejaVu Serif,PrimaryColour=&HCCFF0000'
Scale the input by 2x and smooth using the Super2xSaI (Scale and Interpolate) pixel art scaling algorithm.
Useful for enlarging pixel art images without reducing sharpness.
Swap two rectangular objects in video.
This filter accepts the following options:
Set object width.
Set object height.
Set 1st rect x coordinate.
Set 1st rect y coordinate.
Set 2nd rect x coordinate.
Set 2nd rect y coordinate.
All expressions are evaluated once for each frame.
The all options are expressions containing the following constants:
The input width and height.
same as w / h
input sample aspect ratio
input display aspect ratio, it is the same as (w / h) * sar
The number of the input frame, starting from 0.
The timestamp expressed in seconds. It’s NAN if the input timestamp is unknown.
the position in the file of the input frame, NAN if unknown; deprecated, do not use
This filter supports the all above options as commands.
Swap U & V plane.
Apply telecine process to the video.
This filter accepts the following options:
top field first
bottom field first
The default value is top
.
A string of numbers representing the pulldown pattern you wish to apply.
The default value is 23
.
Some typical patterns: NTSC output (30i): 27.5p: 32222 24p: 23 (classic) 24p: 2332 (preferred) 20p: 33 18p: 334 16p: 3444 PAL output (25i): 27.5p: 12222 24p: 222222222223 ("Euro pulldown") 16.67p: 33 16p: 33333334
Compute and draw a color distribution histogram for the input video across time.
Unlike histogram video filter which only shows histogram of single input frame
at certain time, this filter shows also past histograms of number of frames defined
by width
option.
The computed histogram is a representation of the color component distribution in an image.
The filter accepts the following options:
Set width of single color component output. Default value is 0
.
Value of 0
means width will be picked from input video.
This also set number of passed histograms to keep.
Allowed range is [0, 8192].
Set display mode. It accepts the following values:
Per color component graphs are placed below each other.
Per color component graphs are placed side by side.
Presents information identical to that in the parade
, except
that the graphs representing color components are superimposed directly
over one another.
Default is stack
.
Set mode. Can be either linear
, or logarithmic
.
Default is linear
.
Set what color components to display.
Default is 7
.
Set background opacity. Default is 0.9
.
Show envelope. Default is disabled.
Set envelope color. Default is gold
.
Set slide mode.
Available values for slide is:
Draw new frame when right border is reached.
Replace old columns with new ones.
Scroll from right to left.
Scroll from left to right.
Draw single picture.
Default is replace
.
Apply threshold effect to video stream.
This filter needs four video streams to perform thresholding. First stream is stream we are filtering. Second stream is holding threshold values, third stream is holding min values, and last, fourth stream is holding max values.
The filter accepts the following option:
Set which planes will be processed, unprocessed planes will be copied. By default value 0xf, all planes will be processed.
For example if first stream pixel’s component value is less then threshold value of pixel component from 2nd threshold stream, third stream value will picked, otherwise fourth stream pixel component value will be picked.
Using color source filter one can perform various types of thresholding:
This filter supports the all options as commands.
ffmpeg -i 320x240.avi -f lavfi -i color=gray -f lavfi -i color=black -f lavfi -i color=white -lavfi threshold output.avi
ffmpeg -i 320x240.avi -f lavfi -i color=gray -f lavfi -i color=white -f lavfi -i color=black -lavfi threshold output.avi
ffmpeg -i 320x240.avi -f lavfi -i color=gray -i 320x240.avi -f lavfi -i color=gray -lavfi threshold output.avi
ffmpeg -i 320x240.avi -f lavfi -i color=gray -f lavfi -i color=white -i 320x240.avi -lavfi threshold output.avi
ffmpeg -i 320x240.avi -f lavfi -i color=gray -i 320x240.avi -f lavfi -i color=white -lavfi threshold output.avi
Select the most representative frame in a given sequence of consecutive frames.
The filter accepts the following options:
Set the frames batch size to analyze; in a set of n frames, the filter
will pick one of them, and then handle the next batch of n frames until
the end. Default is 100
.
Set the log level to display picked frame stats.
Default is info
.
Since the filter keeps track of the whole frames sequence, a bigger n value will result in a higher memory usage, so a high value is not recommended.
Tile several successive frames together.
The untile filter can do the reverse.
The filter accepts the following options:
Set the grid size in the form COLUMNSxROWS
. Range is up to UINT_MAX cells.
Default is 6x5
.
Set the maximum number of frames to render in the given area. It must be less
than or equal to wxh. The default value is 0
, meaning all
the area will be used.
Set the outer border margin in pixels. Range is 0 to 1024. Default is 0
.
Set the inner border thickness (i.e. the number of pixels between frames). For
more advanced padding options (such as having different values for the edges),
refer to the pad video filter. Range is 0 to 1024. Default is 0
.
Specify the color of the unused area. For the syntax of this option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual. The default value of color is "black".
Set the number of frames to overlap when tiling several successive frames together.
The value must be between 0
and nb_frames - 1. Default is 0
.
Set the number of frames to initially be empty before displaying first output frame.
This controls how soon will one get first output frame.
The value must be between 0
and nb_frames - 1. Default is 0
.
ffmpeg -skip_frame nokey -i file.avi -vf 'scale=128:72,tile=8x8' -an -vsync 0 keyframes%03d.png
The -vsync 0 is necessary to prevent ffmpeg
from
duplicating each output frame to accommodate the originally detected frame
rate.
5
pictures in an area of 3x2
frames,
with 7
pixels between them, and 2
pixels of initial margin, using
mixed flat and named options:
tile=3x2:nb_frames=5:padding=7:margin=2
Perform various types of temporal field interlacing.
Frames are counted starting from 1, so the first input frame is considered odd.
The filter accepts the following options:
Specify the mode of the interlacing. This option can also be specified as a value alone. See below for a list of values for this option.
Available values are:
Move odd frames into the upper field, even into the lower field, generating a double height frame at half frame rate.
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 11111 33333 22222 44444 11111 33333 22222 44444 11111 33333 22222 44444 11111 33333 22222 44444
Only output odd frames, even frames are dropped, generating a frame with unchanged height at half frame rate.
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 11111 33333 11111 33333 11111 33333 11111 33333
Only output even frames, odd frames are dropped, generating a frame with unchanged height at half frame rate.
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 22222 44444 22222 44444 22222 44444 22222 44444
Expand each frame to full height, but pad alternate lines with black, generating a frame with double height at the same input frame rate.
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 11111 ..... 33333 ..... ..... 22222 ..... 44444 11111 ..... 33333 ..... ..... 22222 ..... 44444 11111 ..... 33333 ..... ..... 22222 ..... 44444 11111 ..... 33333 ..... ..... 22222 ..... 44444
Interleave the upper field from odd frames with the lower field from even frames, generating a frame with unchanged height at half frame rate.
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111<- 22222 33333<- 44444 11111 22222<- 33333 44444<- 11111<- 22222 33333<- 44444 11111 22222<- 33333 44444<- Output: 11111 33333 22222 44444 11111 33333 22222 44444
Interleave the lower field from odd frames with the upper field from even frames, generating a frame with unchanged height at half frame rate.
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222<- 33333 44444<- 11111<- 22222 33333<- 44444 11111 22222<- 33333 44444<- 11111<- 22222 33333<- 44444 Output: 22222 44444 11111 33333 22222 44444 11111 33333
Double frame rate with unchanged height. Frames are inserted each containing the second temporal field from the previous input frame and the first temporal field from the next input frame. This mode relies on the top_field_first flag. Useful for interlaced video displays with no field synchronisation.
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 11111 22222 22222 33333 33333 44444 44444 11111 11111 22222 22222 33333 33333 44444 11111 22222 22222 33333 33333 44444 44444 11111 11111 22222 22222 33333 33333 44444
Move odd frames into the upper field, even into the lower field, generating a double height frame at same frame rate.
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 11111 33333 33333 55555 22222 22222 44444 44444 11111 33333 33333 55555 22222 22222 44444 44444 11111 33333 33333 55555 22222 22222 44444 44444 11111 33333 33333 55555 22222 22222 44444 44444
Numeric values are deprecated but are accepted for backward compatibility reasons.
Default mode is merge
.
Specify flags influencing the filter process.
Available value for flags is:
Enable linear vertical low-pass filtering in the filter. Vertical low-pass filtering is required when creating an interlaced destination from a progressive source which contains high-frequency vertical detail. Filtering will reduce interlace ’twitter’ and Moire patterning.
Enable complex vertical low-pass filtering. This will slightly less reduce interlace ’twitter’ and Moire patterning but better retain detail and subjective sharpness impression.
Bypass already interlaced frames, only adjust the frame rate.
Vertical low-pass filtering and bypassing already interlaced frames can only be enabled for mode interleave_top and interleave_bottom.
Pick median pixels from several successive input video frames.
The filter accepts the following options:
Set radius of median filter. Default is 1. Allowed range is from 1 to 127.
Set which planes to filter. Default value is 15
, by which all planes are processed.
Set median percentile. Default value is 0.5
.
Default value of 0.5
will pick always median values, while 0
will pick
minimum values, and 1
maximum values.
This filter supports all above options as commands, excluding option radius
.
Apply Temporal Midway Video Equalization effect.
Midway Video Equalization adjusts a sequence of video frames to have the same histograms, while maintaining their dynamics as much as possible. It’s useful for e.g. matching exposures from a video frames sequence.
This filter accepts the following option:
Set filtering radius. Default is 5
. Allowed range is from 1 to 127.
Set filtering sigma. Default is 0.5
. This controls strength of filtering.
Setting this option to 0 effectively does nothing.
Set which planes to process. Default is 15
, which is all available planes.
Mix successive video frames.
A description of the accepted options follows.
The number of successive frames to mix. If unspecified, it defaults to 3.
Specify weight of each input video frame. Each weight is separated by space. If number of weights is smaller than number of frames last specified weight will be used for all remaining unset weights.
Specify scale, if it is set it will be multiplied with sum of each weight multiplied with pixel values to give final destination pixel value. By default scale is auto scaled to sum of weights.
Set which planes to filter. Default is all. Allowed range is from 0 to 15.
tmix=frames=7:weights="1 1 1 1 1 1 1"
tmix=frames=3:weights="-1 3 -1"
tmix=frames=3:weights="-1 2 -1":scale=1
Tone map colors from different dynamic ranges.
This filter expects data in single precision floating point, as it needs to operate on (and can output) out-of-range values. Another filter, such as zscale, is needed to convert the resulting frame to a usable format.
The tonemapping algorithms implemented only work on linear light, so input data should be linearized beforehand (and possibly correctly tagged).
ffmpeg -i INPUT -vf zscale=transfer=linear,tonemap=clip,zscale=transfer=bt709,format=yuv420p OUTPUT
The filter accepts the following options.
Set the tone map algorithm to use.
Possible values are:
Do not apply any tone map, only desaturate overbright pixels.
Hard-clip any out-of-range values. Use it for perfect color accuracy for in-range values, while distorting out-of-range values.
Stretch the entire reference gamut to a linear multiple of the display.
Fit a logarithmic transfer between the tone curves.
Preserve overall image brightness with a simple curve, using nonlinear contrast, which results in flattening details and degrading color accuracy.
Preserve both dark and bright details better than reinhard, at the cost of slightly darkening everything. Use it when detail preservation is more important than color and brightness accuracy.
Smoothly map out-of-range values, while retaining contrast and colors for in-range material as much as possible. Use it when color accuracy is more important than detail preservation.
Default is none.
Tune the tone mapping algorithm.
This affects the following algorithms:
Ignored.
Specifies the scale factor to use while stretching. Default to 1.0.
Specifies the exponent of the function. Default to 1.8.
Specify an extra linear coefficient to multiply into the signal before clipping. Default to 1.0.
Specify the local contrast coefficient at the display peak. Default to 0.5, which means that in-gamut values will be about half as bright as when clipping.
Ignored.
Specify the transition point from linear to mobius transform. Every value below this point is guaranteed to be mapped 1:1. The higher the value, the more accurate the result will be, at the cost of losing bright details. Default to 0.3, which due to the steep initial slope still preserves in-range colors fairly accurately.
Apply desaturation for highlights that exceed this level of brightness. The higher the parameter, the more color information will be preserved. This setting helps prevent unnaturally blown-out colors for super-highlights, by (smoothly) turning into white instead. This makes images feel more natural, at the cost of reducing information about out-of-range colors.
The default of 2.0 is somewhat conservative and will mostly just apply to skies or directly sunlit surfaces. A setting of 0.0 disables this option.
This option works only if the input frame has a supported color tag.
Override signal/nominal/reference peak with this value. Useful when the embedded peak information in display metadata is not reliable or when tone mapping from a lower range to a higher range.
Temporarily pad video frames.
The filter accepts the following options:
Specify number of delay frames before input video stream. Default is 0.
Specify number of padding frames after input video stream. Set to -1 to pad indefinitely. Default is 0.
Set kind of frames added to beginning of stream. Can be either add or clone. With add frames of solid-color are added. With clone frames are clones of first frame. Default is add.
Set kind of frames added to end of stream. Can be either add or clone. With add frames of solid-color are added. With clone frames are clones of last frame. Default is add.
Specify the duration of the start/stop delay. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax. These options override start and stop. Default is 0.
Specify the color of the padded area. For the syntax of this option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual.
The default value of color is "black".
Transpose rows with columns in the input video and optionally flip it.
It accepts the following parameters:
Specify the transposition direction.
Can assume the following values:
Rotate by 90 degrees counterclockwise and vertically flip (default), that is:
L.R L.l . . -> . . l.r R.r
Rotate by 90 degrees clockwise, that is:
L.R l.L . . -> . . l.r r.R
Rotate by 90 degrees counterclockwise, that is:
L.R R.r . . -> . . l.r L.l
Rotate by 90 degrees clockwise and vertically flip, that is:
L.R r.R . . -> . . l.r l.L
For values between 4-7, the transposition is only done if the input
video geometry is portrait and not landscape. These values are
deprecated, the passthrough
option should be used instead.
Numerical values are deprecated, and should be dropped in favor of symbolic constants.
Do not apply the transposition if the input geometry matches the one specified by the specified value. It accepts the following values:
Always apply transposition.
Preserve portrait geometry (when height >= width).
Preserve landscape geometry (when width >= height).
Default value is none
.
For example to rotate by 90 degrees clockwise and preserve portrait layout:
transpose=dir=1:passthrough=portrait
The command above can also be specified as:
transpose=1:portrait
Transpose rows with columns in the input video and optionally flip it. For more in depth examples see the transpose video filter, which shares mostly the same options.
It accepts the following parameters:
Specify the transposition direction.
Can assume the following values:
Rotate by 90 degrees counterclockwise and vertically flip. (default)
Rotate by 90 degrees clockwise.
Rotate by 90 degrees counterclockwise.
Rotate by 90 degrees clockwise and vertically flip.
Do not apply the transposition if the input geometry matches the one specified by the specified value. It accepts the following values:
Always apply transposition. (default)
Preserve portrait geometry (when height >= width).
Preserve landscape geometry (when width >= height).
Trim the input so that the output contains one continuous subpart of the input.
It accepts the following parameters:
Specify the time of the start of the kept section, i.e. the frame with the timestamp start will be the first frame in the output.
Specify the time of the first frame that will be dropped, i.e. the frame immediately preceding the one with the timestamp end will be the last frame in the output.
This is the same as start, except this option sets the start timestamp in timebase units instead of seconds.
This is the same as end, except this option sets the end timestamp in timebase units instead of seconds.
The maximum duration of the output in seconds.
The number of the first frame that should be passed to the output.
The number of the first frame that should be dropped.
start, end, and duration are expressed as time duration specifications; see (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax.
Note that the first two sets of the start/end options and the duration option look at the frame timestamp, while the _frame variants simply count the frames that pass through the filter. Also note that this filter does not modify the timestamps. If you wish for the output timestamps to start at zero, insert a setpts filter after the trim filter.
If multiple start or end options are set, this filter tries to be greedy and keep all the frames that match at least one of the specified constraints. To keep only the part that matches all the constraints at once, chain multiple trim filters.
The defaults are such that all the input is kept. So it is possible to set e.g. just the end values to keep everything before the specified time.
Examples:
ffmpeg -i INPUT -vf trim=60:120
ffmpeg -i INPUT -vf trim=duration=1
Apply alpha unpremultiply effect to input video stream using first plane of second stream as alpha.
Both streams must have same dimensions and same pixel format.
The filter accepts the following option:
Set which planes will be processed, unprocessed planes will be copied. By default value 0xf, all planes will be processed.
If the format has 1 or 2 components, then luma is bit 0. If the format has 3 or 4 components: for RGB formats bit 0 is green, bit 1 is blue and bit 2 is red; for YUV formats bit 0 is luma, bit 1 is chroma-U and bit 2 is chroma-V. If present, the alpha channel is always the last bit.
Do not require 2nd input for processing, instead use alpha plane from input stream.
Sharpen or blur the input video.
It accepts the following parameters:
Set the luma matrix horizontal size. It must be an odd integer between 3 and 23. The default value is 5.
Set the luma matrix vertical size. It must be an odd integer between 3 and 23. The default value is 5.
Set the luma effect strength. It must be a floating point number, reasonable values lay between -1.5 and 1.5.
Negative values will blur the input video, while positive values will sharpen it, a value of zero will disable the effect.
Default value is 1.0.
Set the chroma matrix horizontal size. It must be an odd integer between 3 and 23. The default value is 5.
Set the chroma matrix vertical size. It must be an odd integer between 3 and 23. The default value is 5.
Set the chroma effect strength. It must be a floating point number, reasonable values lay between -1.5 and 1.5.
Negative values will blur the input video, while positive values will sharpen it, a value of zero will disable the effect.
Default value is 0.0.
Set the alpha matrix horizontal size. It must be an odd integer between 3 and 23. The default value is 5.
Set the alpha matrix vertical size. It must be an odd integer between 3 and 23. The default value is 5.
Set the alpha effect strength. It must be a floating point number, reasonable values lay between -1.5 and 1.5.
Negative values will blur the input video, while positive values will sharpen it, a value of zero will disable the effect.
Default value is 0.0.
All parameters are optional and default to the equivalent of the string ’5:5:1.0:5:5:0.0’.
Decompose a video made of tiled images into the individual images.
The frame rate of the output video is the frame rate of the input video multiplied by the number of tiles.
This filter does the reverse of tile.
The filter accepts the following options:
Set the grid size (i.e. the number of lines and columns). For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
ffmpeg -r 1 -i image.jpg -vf untile=1x25 movie.mkv
Apply ultra slow/simple postprocessing filter that compresses and decompresses
the image at several (or - in the case of quality level 8
- all)
shifts and average the results.
The way this differs from the behavior of spp is that uspp actually encodes & decodes each case with libavcodec Snow, whereas spp uses a simplified intra only 8x8 DCT similar to MJPEG.
This filter is only available in ffmpeg version 4.4 or earlier.
The filter accepts the following options:
Set quality. This option defines the number of levels for averaging. It accepts
an integer in the range 0-8. If set to 0
, the filter will have no
effect. A value of 8
means the higher quality. For each increment of
that value the speed drops by a factor of approximately 2. Default value is
3
.
Force a constant quantization parameter. If not set, the filter will use the QP from the video stream (if available).
Use specified codec instead of snow.
Convert 360 videos between various formats.
The filter accepts the following options:
Set format of the input/output video.
Available formats:
Equirectangular projection.
Cubemap with 3x2/6x1/1x6 layout.
Format specific options:
Set padding proportion for the input/output cubemap. Values in decimals.
Example values:
No padding.
1% of face is padding. For example, with 1920x1280 resolution face size would be 640x640 and padding would be 3 pixels from each side. (640 * 0.01 = 6 pixels)
Default value is ‘0’. Maximum value is ‘0.1’.
Set fixed padding for the input/output cubemap. Values in pixels.
Default value is ‘0’. If greater than zero it overrides other padding options.
Set order of faces for the input/output cubemap. Choose one direction for each position.
Designation of directions:
right
left
up
down
forward
back
Default value is ‘rludfb’.
Set rotation of faces for the input/output cubemap. Choose one angle for each position.
Designation of angles:
0 degrees clockwise
90 degrees clockwise
180 degrees clockwise
270 degrees clockwise
Default value is ‘000000’.
Equi-Angular Cubemap.
Regular video.
Format specific options:
Set output horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Set input horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Dual fisheye.
Format specific options:
Set output horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Set input horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Facebook’s 360 formats.
Stereographic format.
Format specific options:
Set output horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Set input horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Mercator format.
Ball format, gives significant distortion toward the back.
Hammer-Aitoff map projection format.
Sinusoidal map projection format.
Fisheye projection.
Format specific options:
Set output horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Set input horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Pannini projection.
Format specific options:
Set output pannini parameter.
Set input pannini parameter.
Cylindrical projection.
Format specific options:
Set output horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Set input horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Perspective projection. (output only)
Format specific options:
Set perspective parameter.
Tetrahedron projection.
Truncated square pyramid projection.
Half equirectangular projection.
Equisolid format.
Format specific options:
Set output horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Set input horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Orthographic format.
Format specific options:
Set output horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Set input horizontal/vertical/diagonal field of view. Values in degrees.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Octahedron projection.
Cylindrical Equal Area projection.
Set interpolation method.
Note: more complex interpolation methods require much more memory to run.
Available methods:
Nearest neighbour.
Bilinear interpolation.
Lagrange9 interpolation.
Bicubic interpolation.
Lanczos interpolation.
Spline16 interpolation.
Gaussian interpolation.
Mitchell interpolation.
Default value is ‘line’.
Set the output video resolution.
Default resolution depends on formats.
Set the input/output stereo format.
2D mono
Side by side
Top bottom
Default value is ‘2d’ for input and output format.
Set rotation for the output video. Values in degrees.
Set rotation order for the output video. Choose one item for each position.
yaw
pitch
roll
Default value is ‘ypr’.
Flip the output video horizontally(swaps left-right)/vertically(swaps up-down)/in-depth(swaps back-forward). Boolean values.
Set if input video is flipped horizontally/vertically. Boolean values.
Set if input video is transposed. Boolean value, by default disabled.
Set if output video needs to be transposed. Boolean value, by default disabled.
Set output horizontal/vertical off-axis offset. Default is set to 0. Allowed range is from -1 to 1.
Build mask in alpha plane for all unmapped pixels by marking them fully transparent. Boolean value, by default disabled.
Reset rotation of output video. Boolean value, by default disabled.
ffmpeg -i input.mkv -vf v360=e:c3x2:cubic:out_pad=0.01 output.mkv
ffmpeg -i input.mkv -vf v360=eac:flat:yaw=180 output.mkv
v360=eac:equirect:in_stereo=sbs:in_trans=1:ih_flip=1:out_stereo=tb
This filter supports subset of above options as commands.
Apply a wavelet based denoiser.
It transforms each frame from the video input into the wavelet domain, using Cohen-Daubechies-Feauveau 9/7. Then it applies some filtering to the obtained coefficients. It does an inverse wavelet transform after. Due to wavelet properties, it should give a nice smoothed result, and reduced noise, without blurring picture features.
This filter accepts the following options:
The filtering strength. The higher, the more filtered the video will be. Hard thresholding can use a higher threshold than soft thresholding before the video looks overfiltered. Default value is 2.
The filtering method the filter will use.
It accepts the following values:
All values under the threshold will be zeroed.
All values under the threshold will be zeroed. All values above will be reduced by the threshold.
Scales or nullifies coefficients - intermediary between (more) soft and (less) hard thresholding.
Default is garrote.
Number of times, the wavelet will decompose the picture. Picture can’t be decomposed beyond a particular point (typically, 8 for a 640x480 frame - as 2^9 = 512 > 480). Valid values are integers between 1 and 32. Default value is 6.
Partial of full denoising (limited coefficients shrinking), from 0 to 100. Default value is 85.
A list of the planes to process. By default all planes are processed.
The threshold type the filter will use.
It accepts the following values:
Threshold used is same for all decompositions.
Threshold used depends also on each decomposition coefficients.
Default is universal.
Apply variable blur filter by using 2nd video stream to set blur radius. The 2nd stream must have the same dimensions.
This filter accepts the following options:
Set min allowed radius. Allowed range is from 0 to 254. Default is 0.
Set max allowed radius. Allowed range is from 1 to 255. Default is 8.
Set which planes to process. By default, all are used.
The varblur
filter also supports the framesync options.
This filter supports all the above options as commands.
Display 2 color component values in the two dimensional graph (which is called a vectorscope).
This filter accepts the following options:
Set vectorscope mode.
It accepts the following values:
Gray values are displayed on graph, higher brightness means more pixels have same component color value on location in graph. This is the default mode.
Gray values are displayed on graph. Surrounding pixels values which are not
present in video frame are drawn in gradient of 2 color components which are
set by option x
and y
. The 3rd color component is static.
Actual color components values present in video frame are displayed on graph.
Similar as color2 but higher frequency of same values x
and y
on graph increases value of another color component, which is luminance by
default values of x
and y
.
Actual colors present in video frame are displayed on graph. If two different colors map to same position on graph then color with higher value of component not present in graph is picked.
Gray values are displayed on graph. Similar to color
but with 3rd color
component picked from radial gradient.
Set which color component will be represented on X-axis. Default is 1
.
Set which color component will be represented on Y-axis. Default is 2
.
Set intensity, used by modes: gray, color, color3 and color5 for increasing brightness of color component which represents frequency of (X, Y) location in graph.
No envelope, this is default.
Instant envelope, even darkest single pixel will be clearly highlighted.
Hold maximum and minimum values presented in graph over time. This way you can still spot out of range values without constantly looking at vectorscope.
Peak and instant envelope combined together.
Set what kind of graticule to draw.
Set graticule opacity.
Set graticule flags.
Draw graticule for white point.
Draw graticule for black point.
Draw color points short names.
Set background opacity.
Set low threshold for color component not represented on X or Y axis. Values lower than this value will be ignored. Default is 0. Note this value is multiplied with actual max possible value one pixel component can have. So for 8-bit input and low threshold value of 0.1 actual threshold is 0.1 * 255 = 25.
Set high threshold for color component not represented on X or Y axis. Values higher than this value will be ignored. Default is 1. Note this value is multiplied with actual max possible value one pixel component can have. So for 8-bit input and high threshold value of 0.9 actual threshold is 0.9 * 255 = 230.
Set what kind of colorspace to use when drawing graticule.
Default is auto.
Set color tint for gray/tint vectorscope mode. By default both options are zero. This means no tint, and output will remain gray.
Analyze video stabilization/deshaking. Perform pass 1 of 2, see vidstabtransform for pass 2.
This filter generates a file with relative translation and rotation transform information about subsequent frames, which is then used by the vidstabtransform filter.
To enable compilation of this filter you need to configure FFmpeg with
--enable-libvidstab
.
This filter accepts the following options:
Set the path to the file used to write the transforms information. Default value is transforms.trf.
Set how shaky the video is and how quick the camera is. It accepts an integer in the range 1-10, a value of 1 means little shakiness, a value of 10 means strong shakiness. Default value is 5.
Set the accuracy of the detection process. It must be a value in the range 1-15. A value of 1 means low accuracy, a value of 15 means high accuracy. Default value is 15.
Set stepsize of the search process. The region around minimum is scanned with 1 pixel resolution. Default value is 6.
Set minimum contrast. Below this value a local measurement field is discarded. Must be a floating point value in the range 0-1. Default value is 0.3.
Set reference frame number for tripod mode.
If enabled, the motion of the frames is compared to a reference frame in the filtered stream, identified by the specified number. The idea is to compensate all movements in a more-or-less static scene and keep the camera view absolutely still.
If set to 0, it is disabled. The frames are counted starting from 1.
Show fields and transforms in the resulting frames. It accepts an integer in the range 0-2. Default value is 0, which disables any visualization.
Format for the transforms data file to be written. Acceptable values are
Human-readable plain text
Binary format, roughly 40% smaller than ascii
. (default)
vidstabdetect
vidstabdetect=shakiness=10:accuracy=15:result="mytransforms.trf"
vidstabdetect=show=1
ffmpeg
:
ffmpeg -i input -vf vidstabdetect=shakiness=5:show=1 dummy.avi
Video stabilization/deshaking: pass 2 of 2, see vidstabdetect for pass 1.
Read a file with transform information for each frame and apply/compensate them. Together with the vidstabdetect filter this can be used to deshake videos. See also http://public.hronopik.de/vid.stab. It is important to also use the unsharp filter, see below.
To enable compilation of this filter you need to configure FFmpeg with
--enable-libvidstab
.
Set path to the file used to read the transforms. Default value is transforms.trf.
Set the number of frames (value*2 + 1) used for lowpass filtering the camera movements. Default value is 10.
For example a number of 10 means that 21 frames are used (10 in the past and 10 in the future) to smoothen the motion in the video. A larger value leads to a smoother video, but limits the acceleration of the camera (pan/tilt movements). 0 is a special case where a static camera is simulated.
Set the camera path optimization algorithm.
Accepted values are:
gaussian kernel low-pass filter on camera motion (default)
averaging on transformations
Set maximal number of pixels to translate frames. Default value is -1, meaning no limit.
Set maximal angle in radians (degree*PI/180) to rotate frames. Default value is -1, meaning no limit.
Specify how to deal with borders that may be visible due to movement compensation.
Available values are:
keep image information from previous frame (default)
fill the border black
Invert transforms if set to 1. Default value is 0.
Consider transforms as relative to previous frame if set to 1, absolute if set to 0. Default value is 0.
Set percentage to zoom. A positive value will result in a zoom-in effect, a negative value in a zoom-out effect. Default value is 0 (no zoom).
Set optimal zooming to avoid borders.
Accepted values are:
disabled
optimal static zoom value is determined (only very strong movements will lead to visible borders) (default)
optimal adaptive zoom value is determined (no borders will be visible), see zoomspeed
Note that the value given at zoom is added to the one calculated here.
Set percent to zoom maximally each frame (enabled when optzoom is set to 2). Range is from 0 to 5, default value is 0.25.
Specify type of interpolation.
Available values are:
no interpolation
linear only horizontal
linear in both directions (default)
cubic in both directions (slow)
Enable virtual tripod mode if set to 1, which is equivalent to
relative=0:smoothing=0
. Default value is 0.
Use also tripod
option of vidstabdetect.
Increase log verbosity if set to 1. Also the detected global motions are written to the temporary file global_motions.trf. Default value is 0.
ffmpeg
for a typical stabilization with default values:
ffmpeg -i inp.mpeg -vf vidstabtransform,unsharp=5:5:0.8:3:3:0.4 inp_stabilized.mpeg
Note the use of the unsharp filter which is always recommended.
vidstabtransform=zoom=5:input="mytransforms.trf"
vidstabtransform=smoothing=30
Flip the input video vertically.
For example, to vertically flip a video with ffmpeg
:
ffmpeg -i in.avi -vf "vflip" out.avi
Detect variable frame rate video.
This filter tries to detect if the input is variable or constant frame rate.
At end it will output number of frames detected as having variable delta pts, and ones with constant delta pts. If there was frames with variable delta, than it will also show min, max and average delta encountered.
Boost or alter saturation.
The filter accepts the following options:
Set strength of boost if positive value or strength of alter if negative value. Default is 0. Allowed range is from -2 to 2.
Set the red balance. Default is 1. Allowed range is from -10 to 10.
Set the green balance. Default is 1. Allowed range is from -10 to 10.
Set the blue balance. Default is 1. Allowed range is from -10 to 10.
Set the red luma coefficient.
Set the green luma coefficient.
Set the blue luma coefficient.
If intensity
is negative and this is set to 1, colors will change,
otherwise colors will be less saturated, more towards gray.
This filter supports the all above options as commands.
Obtain the average VIF (Visual Information Fidelity) between two input videos.
This filter takes two input videos.
Both input videos must have the same resolution and pixel format for this filter to work correctly. Also it assumes that both inputs have the same number of frames, which are compared one by one.
The obtained average VIF score is printed through the logging system.
The filter stores the calculated VIF score of each frame.
This filter also supports the framesync options.
In the below example the input file main.mpg being processed is compared with the reference file ref.mpg.
ffmpeg -i main.mpg -i ref.mpg -lavfi vif -f null -
Make or reverse a natural vignetting effect.
The filter accepts the following options:
Set lens angle expression as a number of radians.
The value is clipped in the [0,PI/2]
range.
Default value: "PI/5"
Set center coordinates expressions. Respectively "w/2"
and "h/2"
by default.
Set forward/backward mode.
Available modes are:
The larger the distance from the central point, the darker the image becomes.
The larger the distance from the central point, the brighter the image becomes. This can be used to reverse a vignette effect, though there is no automatic detection to extract the lens angle and other settings (yet). It can also be used to create a burning effect.
Default value is ‘forward’.
Set evaluation mode for the expressions (angle, x0, y0).
It accepts the following values:
Evaluate expressions only once during the filter initialization.
Evaluate expressions for each incoming frame. This is way slower than the ‘init’ mode since it requires all the scalers to be re-computed, but it allows advanced dynamic expressions.
Default value is ‘init’.
Set dithering to reduce the circular banding effects. Default is 1
(enabled).
Set vignette aspect. This setting allows one to adjust the shape of the vignette. Setting this value to the SAR of the input will make a rectangular vignetting following the dimensions of the video.
Default is 1/1
.
The alpha, x0 and y0 expressions can contain the following parameters.
input width and height
the number of input frame, starting from 0
the PTS (Presentation TimeStamp) time of the filtered video frame, expressed in TB units, NAN if undefined
frame rate of the input video, NAN if the input frame rate is unknown
the PTS (Presentation TimeStamp) of the filtered video frame, expressed in seconds, NAN if undefined
time base of the input video
vignette=PI/4
vignette='PI/4+random(1)*PI/50':eval=frame
Obtain the average VMAF motion score of a video. It is one of the component metrics of VMAF.
The obtained average motion score is printed through the logging system.
The filter accepts the following options:
If specified, the filter will use the named file to save the motion score of each frame with respect to the previous frame. When filename equals "-" the data is sent to standard output.
Example:
ffmpeg -i ref.mpg -vf vmafmotion -f null -
Stack input videos vertically.
All streams must be of same pixel format and of same width.
Note that this filter is faster than using overlay and pad filter to create same output.
The filter accepts the following options:
Set number of input streams. Default is 2.
If set to 1, force the output to terminate when the shortest input terminates. Default value is 0.
Deinterlace the input video ("w3fdif" stands for "Weston 3 Field Deinterlacing Filter").
Based on the process described by Martin Weston for BBC R&D, and implemented based on the de-interlace algorithm written by Jim Easterbrook for BBC R&D, the Weston 3 field deinterlacing filter uses filter coefficients calculated by BBC R&D.
This filter uses field-dominance information in frame to decide which
of each pair of fields to place first in the output.
If it gets it wrong use setfield filter before w3fdif
filter.
There are two sets of filter coefficients, so called "simple" and "complex". Which set of filter coefficients is used can be set by passing an optional parameter:
Set the interlacing filter coefficients. Accepts one of the following values:
Simple filter coefficient set.
More-complex filter coefficient set.
Default value is ‘complex’.
The interlacing mode to adopt. It accepts one of the following values:
Output one frame for each frame.
Output one frame for each field.
The default value is field
.
The picture field parity assumed for the input interlaced video. It accepts one of the following values:
Assume the top field is first.
Assume the bottom field is first.
Enable automatic detection of field parity.
The default value is auto
.
If the interlacing is unknown or the decoder does not export this information,
top field first will be assumed.
Specify which frames to deinterlace. Accepts one of the following values:
Deinterlace all frames,
Only deinterlace frames marked as interlaced.
Default value is ‘all’.
This filter supports same commands as options.
Video waveform monitor.
The waveform monitor plots color component intensity. By default luma only. Each column of the waveform corresponds to a column of pixels in the source video.
It accepts the following options:
Can be either row
, or column
. Default is column
.
In row mode, the graph on the left side represents color component value 0 and
the right side represents value = 255. In column mode, the top side represents
color component value = 0 and bottom side represents value = 255.
Set intensity. Smaller values are useful to find out how many values of the same
luminance are distributed across input rows/columns.
Default value is 0.04
. Allowed range is [0, 1].
Set mirroring mode. 0
means unmirrored, 1
means mirrored.
In mirrored mode, higher values will be represented on the left
side for row
mode and at the top for column
mode. Default is
1
(mirrored).
Set display mode. It accepts the following values:
Presents information identical to that in the parade
, except
that the graphs representing color components are superimposed directly
over one another.
This display mode makes it easier to spot relative differences or similarities in overlapping areas of the color components that are supposed to be identical, such as neutral whites, grays, or blacks.
Display separate graph for the color components side by side in
row
mode or one below the other in column
mode.
Display separate graph for the color components side by side in
column
mode or one below the other in row
mode.
Using this display mode makes it easy to spot color casts in the highlights and shadows of an image, by comparing the contours of the top and the bottom graphs of each waveform. Since whites, grays, and blacks are characterized by exactly equal amounts of red, green, and blue, neutral areas of the picture should display three waveforms of roughly equal width/height. If not, the correction is easy to perform by making level adjustments the three waveforms.
Default is stack
.
Set which color components to display. Default is 1, which means only luma or red color component if input is in RGB colorspace. If is set for example to 7 it will display all 3 (if) available color components.
No envelope, this is default.
Instant envelope, minimum and maximum values presented in graph will be easily
visible even with small step
value.
Hold minimum and maximum values presented in graph across time. This way you can still spot out of range values without constantly looking at waveforms.
Peak and instant envelope combined together.
No filtering, this is default.
Luma and chroma combined together.
Similar as above, but shows difference between blue and red chroma.
Similar as above, but use different colors.
Similar as above, but again with different colors.
Displays only chroma.
Displays actual color value on waveform.
Similar as above, but with luma showing frequency of chroma values.
Set which graticule to display.
Do not display graticule.
Display green graticule showing legal broadcast ranges.
Display orange graticule showing legal broadcast ranges.
Display invert graticule showing legal broadcast ranges.
Set graticule opacity.
Set graticule flags.
Draw numbers above lines. By default enabled.
Draw dots instead of lines.
Set scale used for displaying graticule.
Default is digital.
Set background opacity.
Set tint for output. Only used with lowpass filter and when display is not overlay and input pixel formats are not RGB.
Set sample aspect ratio of video output frames. Can be used to configure waveform so it is not streched too much in one of directions.
Set sample aspect ration to 1/1.
Set sample aspect ratio to match input size of video
Default is ‘none’.
Set input formats for filter to pick from. Can be ‘all’, for selecting from all available formats, or ‘first’, for selecting first available format. Default is ‘first’.
The weave
takes a field-based video input and join
each two sequential fields into single frame, producing a new double
height clip with half the frame rate and half the frame count.
The doubleweave
works same as weave
but without
halving frame rate and frame count.
It accepts the following option:
Set first field. Available values are:
Set the frame as top-field-first.
Set the frame as bottom-field-first.
separatefields,select=eq(mod(n,4),0)+eq(mod(n,4),3),weave
Apply the xBR high-quality magnification filter which is designed for pixel art. It follows a set of edge-detection rules, see https://forums.libretro.com/t/xbr-algorithm-tutorial/123.
It accepts the following option:
Set the scaling dimension: 2
for 2xBR
, 3
for
3xBR
and 4
for 4xBR
.
Default is 3
.
Apply normalized cross-correlation between first and second input video stream.
Second input video stream dimensions must be lower than first input video stream.
The filter accepts the following options:
Set which planes to process.
Set which secondary video frames will be processed from second input video stream, can be first or all. Default is all.
The xcorrelate
filter also supports the framesync options.
Apply cross fade from one input video stream to another input video stream. The cross fade is applied for specified duration.
Both inputs must be constant frame-rate and have the same resolution, pixel format, frame rate and timebase.
The filter accepts the following options:
Set one of available transition effects:
Default transition effect is fade.
Set cross fade duration in seconds. Range is 0 to 60 seconds. Default duration is 1 second.
Set cross fade start relative to first input stream in seconds. Default offset is 0.
Set expression for custom transition effect.
The expressions can use the following variables and functions:
The coordinates of the current sample.
The width and height of the image.
Progress of transition effect.
Currently processed plane.
Return value of first input at current location and plane.
Return value of second input at current location and plane.
Return the value of the pixel at location (x,y) of the first/second/third/fourth component of first input.
Return the value of the pixel at location (x,y) of the first/second/third/fourth component of second input.
ffmpeg -i first.mp4 -i second.mp4 -filter_complex xfade=transition=fade:duration=2:offset=5 output.mp4
Pick median pixels from several input videos.
The filter accepts the following options:
Set number of inputs. Default is 3. Allowed range is from 3 to 255. If number of inputs is even number, than result will be mean value between two median values.
Set which planes to filter. Default value is 15
, by which all planes are processed.
Set median percentile. Default value is 0.5
.
Default value of 0.5
will pick always median values, while 0
will pick
minimum values, and 1
maximum values.
Stack video inputs into custom layout.
All streams must be of same pixel format.
The filter accepts the following options:
Set number of input streams. Default is 2.
Specify layout of inputs. This option requires the desired layout configuration to be explicitly set by the user. This sets position of each video input in output. Each input is separated by ’|’. The first number represents the column, and the second number represents the row. Numbers start at 0 and are separated by ’_’. Optionally one can use wX and hX, where X is video input from which to take width or height. Multiple values can be used when separated by ’+’. In such case values are summed together.
Note that if inputs are of different sizes gaps may appear, as not all of the output video frame will be filled. Similarly, videos can overlap each other if their position doesn’t leave enough space for the full frame of adjoining videos.
For 2 inputs, a default layout of 0_0|w0_0
(equivalent to
grid=2x1
) is set. In all other cases, a layout or a grid must be set by
the user. Either grid
or layout
can be specified at a time.
Specifying both will result in an error.
Specify a fixed size grid of inputs.
This option is used to create a fixed size grid of the input streams. Set the
grid size in the form COLUMNSxROWS
. There must be ROWS * COLUMNS
input streams and they will be arranged as a grid with ROWS
rows and
COLUMNS
columns. When using this option, each input stream within a row
must have the same height and all the rows must have the same width.
If grid
is set, then inputs
option is ignored and is implicitly
set to ROWS * COLUMNS
.
For 2 inputs, a default grid of 2x1
(equivalent to
layout=0_0|w0_0
) is set. In all other cases, a layout or a grid must be
set by the user. Either grid
or layout
can be specified at a time.
Specifying both will result in an error.
If set to 1, force the output to terminate when the shortest input terminates. Default value is 0.
If set to valid color, all unused pixels will be filled with that color. By default fill is set to none, so it is disabled.
Layout:
input1(0, 0) | input3(w0, 0) input2(0, h0) | input4(w0, h0)
xstack=inputs=4:layout=0_0|0_h0|w0_0|w0_h0
Note that if inputs are of different sizes, gaps or overlaps may occur.
Layout:
input1(0, 0) input2(0, h0) input3(0, h0+h1) input4(0, h0+h1+h2)
xstack=inputs=4:layout=0_0|0_h0|0_h0+h1|0_h0+h1+h2
Note that if inputs are of different widths, unused space will appear.
Layout:
input1(0, 0) | input4(w0, 0) | input7(w0+w3, 0) input2(0, h0) | input5(w0, h0) | input8(w0+w3, h0) input3(0, h0+h1) | input6(w0, h0+h1) | input9(w0+w3, h0+h1)
xstack=inputs=9:layout=0_0|0_h0|0_h0+h1|w0_0|w0_h0|w0_h0+h1|w0+w3_0|w0+w3_h0|w0+w3_h0+h1
Note that if inputs are of different sizes, gaps or overlaps may occur.
Layout:
input1(0, 0) | input5(w0, 0) | input9 (w0+w4, 0) | input13(w0+w4+w8, 0) input2(0, h0) | input6(w0, h0) | input10(w0+w4, h0) | input14(w0+w4+w8, h0) input3(0, h0+h1) | input7(w0, h0+h1) | input11(w0+w4, h0+h1) | input15(w0+w4+w8, h0+h1) input4(0, h0+h1+h2)| input8(w0, h0+h1+h2)| input12(w0+w4, h0+h1+h2)| input16(w0+w4+w8, h0+h1+h2)
xstack=inputs=16:layout=0_0|0_h0|0_h0+h1|0_h0+h1+h2|w0_0|w0_h0|w0_h0+h1|w0_h0+h1+h2|w0+w4_0| w0+w4_h0|w0+w4_h0+h1|w0+w4_h0+h1+h2|w0+w4+w8_0|w0+w4+w8_h0|w0+w4+w8_h0+h1|w0+w4+w8_h0+h1+h2
Note that if inputs are of different sizes, gaps or overlaps may occur.
Deinterlace the input video ("yadif" means "yet another deinterlacing filter").
It accepts the following parameters:
The interlacing mode to adopt. It accepts one of the following values:
Output one frame for each frame.
Output one frame for each field.
Like send_frame
, but it skips the spatial interlacing check.
Like send_field
, but it skips the spatial interlacing check.
The default value is send_frame
.
The picture field parity assumed for the input interlaced video. It accepts one of the following values:
Assume the top field is first.
Assume the bottom field is first.
Enable automatic detection of field parity.
The default value is auto
.
If the interlacing is unknown or the decoder does not export this information,
top field first will be assumed.
Specify which frames to deinterlace. Accepts one of the following values:
Deinterlace all frames.
Only deinterlace frames marked as interlaced.
The default value is all
.
Deinterlace the input video using the yadif algorithm, but implemented in CUDA so that it can work as part of a GPU accelerated pipeline with nvdec and/or nvenc.
It accepts the following parameters:
The interlacing mode to adopt. It accepts one of the following values:
Output one frame for each frame.
Output one frame for each field.
Like send_frame
, but it skips the spatial interlacing check.
Like send_field
, but it skips the spatial interlacing check.
The default value is send_frame
.
The picture field parity assumed for the input interlaced video. It accepts one of the following values:
Assume the top field is first.
Assume the bottom field is first.
Enable automatic detection of field parity.
The default value is auto
.
If the interlacing is unknown or the decoder does not export this information,
top field first will be assumed.
Specify which frames to deinterlace. Accepts one of the following values:
Deinterlace all frames.
Only deinterlace frames marked as interlaced.
The default value is all
.
Apply blur filter while preserving edges ("yaepblur" means "yet another edge preserving blur filter"). The algorithm is described in "J. S. Lee, Digital image enhancement and noise filtering by use of local statistics, IEEE Trans. Pattern Anal. Mach. Intell. PAMI-2, 1980."
It accepts the following parameters:
Set the window radius. Default value is 3.
Set which planes to filter. Default is only the first plane.
Set blur strength. Default value is 128.
This filter supports same commands as options.
Apply Zoom & Pan effect.
This filter accepts the following options:
Set the zoom expression. Range is 1-10. Default is 1.
Set the x and y expression. Default is 0.
Set the duration expression in number of frames. This sets for how many number of frames effect will last for single input image. Default is 90.
Set the output image size, default is ’hd720’.
Set the output frame rate, default is ’25’.
Each expression can contain the following constants:
Input width.
Input height.
Output width.
Output height.
Input frame count.
Output frame count.
The input timestamp expressed in seconds. It’s NAN if the input timestamp is unknown.
The output timestamp expressed in seconds.
Last calculated ’x’ and ’y’ position from ’x’ and ’y’ expression for current input frame.
’x’ and ’y’ of last output frame of previous input frame or 0 when there was not yet such frame (first input frame).
Last calculated zoom from ’z’ expression for current input frame.
Last calculated zoom of last output frame of previous input frame.
Number of output frames for current input frame. Calculated from ’d’ expression for each input frame.
number of output frames created for previous input frame
Rational number: input width / input height
sample aspect ratio
display aspect ratio
zoompan=z='min(zoom+0.0015,1.5)':d=700:x='if(gte(zoom,1.5),x,x+1/a)':y='if(gte(zoom,1.5),y,y+1)':s=640x360
zoompan=z='min(zoom+0.0015,1.5)':d=700:x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)'
zoompan=z='min(max(zoom,pzoom)+0.0015,1.5)':d=1:x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)'
zoompan=z='if(between(in_time,0,1),2,1)':d=1:x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)'
Scale (resize) the input video, using the z.lib library:
https://github.com/sekrit-twc/zimg. To enable compilation of this
filter, you need to configure FFmpeg with --enable-libzimg
.
The zscale filter forces the output display aspect ratio to be the same as the input, by changing the output sample aspect ratio.
If the input image format is different from the format requested by the next filter, the zscale filter will convert the input to the requested format.
The filter accepts the following options.
Set the output video dimension expression. Default value is the input dimension.
If the width or w value is 0, the input width is used for the output. If the height or h value is 0, the input height is used for the output.
If one and only one of the values is -n with n >= 1, the zscale filter will use a value that maintains the aspect ratio of the input image, calculated from the other specified dimension. After that it will, however, make sure that the calculated dimension is divisible by n and adjust the value if necessary.
If both values are -n with n >= 1, the behavior will be identical to both values being set to 0 as previously detailed.
See below for the list of accepted constants for use in the dimension expression.
Set the video size. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Set the dither type.
Possible values are:
Default is none.
Set the resize filter type.
Possible values are:
Default is bilinear.
Set the color range.
Possible values are:
Default is same as input.
Set the color primaries.
Possible values are:
Default is same as input.
Set the transfer characteristics.
Possible values are:
Default is same as input.
Set the colorspace matrix.
Possible value are:
Default is same as input.
Set the input color range.
Possible values are:
Default is same as input.
Set the input color primaries.
Possible values are:
Default is same as input.
Set the input transfer characteristics.
Possible values are:
Default is same as input.
Set the input colorspace matrix.
Possible value are:
Set the output chroma location.
Possible values are:
Set the input chroma location.
Possible values are:
Set the nominal peak luminance.
Parameter A for scaling filters. Parameter "b" for bicubic, and the number of filter taps for lanczos.
Parameter B for scaling filters. Parameter "c" for bicubic.
The values of the w and h options are expressions containing the following constants:
The input width and height
These are the same as in_w and in_h.
The output (scaled) width and height
These are the same as out_w and out_h
The same as iw / ih
input sample aspect ratio
The input display aspect ratio. Calculated from (iw / ih) * sar
.
horizontal and vertical input chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
horizontal and vertical output chroma subsample values. For example for the pixel format "yuv422p" hsub is 2 and vsub is 1.
This filter supports the following commands:
Set the output video dimension expression. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Below is a description of the currently available OpenCL video filters.
To enable compilation of these filters you need to configure FFmpeg with
--enable-opencl
.
Running OpenCL filters requires you to initialize a hardware device and to pass that device to all filters in any filter graph.
Initialise a new hardware device of type opencl called name, using the given device parameters.
Pass the hardware device called name to all filters in any filter graph.
For more detailed information see https://www.ffmpeg.org/ffmpeg.html#Advanced-Video-options
-init_hw_device opencl=gpu:1.0 -filter_hw_device gpu -i INPUT -vf "hwupload, avgblur_opencl, hwdownload" OUTPUT
Since OpenCL filters are not able to access frame data in normal memory, all frame data needs to be uploaded(hwupload) to hardware surfaces connected to the appropriate device before being used and then downloaded(hwdownload) back to normal memory. Note that hwupload will upload to a surface with the same layout as the software frame, so it may be necessary to add a format filter immediately before to get the input into the right format and hwdownload does not support all formats on the output - it may be necessary to insert an additional format filter immediately following in the graph to get the output in a supported format.
Apply average blur filter.
The filter accepts the following options:
Set horizontal radius size.
Range is [1, 1024]
and default value is 1
.
Set which planes to filter. Default value is 0xf
, by which all planes are processed.
Set vertical radius size. Range is [1, 1024]
and default value is 0
. If zero, sizeX
value will be used.
-i INPUT -vf "hwupload, avgblur_opencl=3, hwdownload" OUTPUT
Apply a boxblur algorithm to the input video.
It accepts the following parameters:
A description of the accepted options follows.
Set an expression for the box radius in pixels used for blurring the corresponding input plane.
The radius value must be a non-negative number, and must not be
greater than the value of the expression min(w,h)/2
for the
luma and alpha planes, and of min(cw,ch)/2
for the chroma
planes.
Default value for luma_radius is "2". If not specified, chroma_radius and alpha_radius default to the corresponding value set for luma_radius.
The expressions can contain the following constants:
The input width and height in pixels.
The input chroma image width and height in pixels.
The horizontal and vertical chroma subsample values. For example, for the pixel format "yuv422p", hsub is 2 and vsub is 1.
Specify how many times the boxblur filter is applied to the corresponding plane.
Default value for luma_power is 2. If not specified, chroma_power and alpha_power default to the corresponding value set for luma_power.
A value of 0 will disable the effect.
Apply boxblur filter, setting each pixel of the output to the average value of box-radiuses luma_radius, chroma_radius, alpha_radius for each plane respectively. The filter will apply luma_power, chroma_power, alpha_power times onto the corresponding plane. For pixels on the edges of the image, the radius does not extend beyond the image boundaries, and so out-of-range coordinates are not used in the calculations.
-i INPUT -vf "hwupload, boxblur_opencl=luma_radius=2:luma_power=3, hwdownload" OUTPUT -i INPUT -vf "hwupload, boxblur_opencl=2:3, hwdownload" OUTPUT
For the luma plane, a 2x2 box radius will be run once.
For the chroma plane, a 4x4 box radius will be run 5 times.
For the alpha plane, a 3x3 box radius will be run 7 times.
-i INPUT -vf "hwupload, boxblur_opencl=2:1:4:5:3:7, hwdownload" OUTPUT
RGB colorspace color keying.
The filter accepts the following options:
The color which will be replaced with transparency.
Similarity percentage with the key color.
0.01 matches only the exact key color, while 1.0 matches everything.
Blend percentage.
0.0 makes pixels either fully transparent, or not transparent at all.
Higher values result in semi-transparent pixels, with a higher transparency the more similar the pixels color is to the key color.
-i INPUT -vf "hwupload, colorkey_opencl=green:0.3:0.1, hwdownload" OUTPUT
Apply convolution of 3x3, 5x5, 7x7 matrix.
The filter accepts the following options:
Set matrix for each plane.
Matrix is sequence of 9, 25 or 49 signed numbers.
Default value for each plane is 0 0 0 0 1 0 0 0 0
.
Set multiplier for calculated value for each plane.
If unset or 0, it will be sum of all matrix elements.
The option value must be a float number greater or equal to 0.0
. Default value is 1.0
.
Set bias for each plane. This value is added to the result of the multiplication.
Useful for making the overall image brighter or darker.
The option value must be a float number greater or equal to 0.0
. Default value is 0.0
.
-i INPUT -vf "hwupload, convolution_opencl=0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0, hwdownload" OUTPUT
-i INPUT -vf "hwupload, convolution_opencl=1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1/9:1/9:1/9:1/9, hwdownload" OUTPUT
-i INPUT -vf "hwupload, convolution_opencl=0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:5:1:1:1:0:128:128:128, hwdownload" OUTPUT
-i INPUT -vf "hwupload, convolution_opencl=0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:5:5:5:1:0:128:128:128, hwdownload" OUTPUT
-i INPUT -vf "hwupload, convolution_opencl=1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:5:5:5:1:0:128:128:0, hwdownload" OUTPUT
-i INPUT -vf "hwupload, convolution_opencl=-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2, hwdownload" OUTPUT
Apply erosion effect to the video.
This filter replaces the pixel by the local(3x3) minimum.
It accepts the following options:
Limit the maximum change for each plane. Range is [0, 65535]
and default value is 65535
.
If 0
, plane will remain unchanged.
Flag which specifies the pixel to refer to.
Range is [0, 255]
and default value is 255
, i.e. all eight pixels are used.
Flags to local 3x3 coordinates region centered on x
:
1 2 3
4 x 5
6 7 8
-i INPUT -vf "hwupload, erosion_opencl=30:40:50:coordinates=231, hwdownload" OUTPUT
Feature-point based video stabilization filter.
The filter accepts the following options:
Simulates a tripod by preventing any camera movement whatsoever from the original frame. Defaults to 0
.
Whether or not additional debug info should be displayed, both in the processed output and in the console.
Note that in order to see console debug output you will also need to pass -v verbose
to ffmpeg.
Viewing point matches in the output video is only supported for RGB input.
Defaults to 0
.
Whether or not to do a tiny bit of cropping at the borders to cut down on the amount of mirrored pixels.
Defaults to 1
.
Whether or not feature points should be refined at a sub-pixel level.
This can be turned off for a slight performance gain at the cost of precision.
Defaults to 1
.
The strength of the smoothing applied to the camera path from 0.0
to 1.0
.
1.0
is the maximum smoothing strength while values less than that result in less smoothing.
0.0
causes the filter to adaptively choose a smoothing strength on a per-frame basis.
Defaults to 0.0
.
Controls the size of the smoothing window (the number of frames buffered to determine motion information from).
The size of the smoothing window is determined by multiplying the framerate of the video by this number.
Acceptable values range from 0.1
to 10.0
.
Larger values increase the amount of motion data available for determining how to smooth the camera path, potentially improving smoothness, but also increase latency and memory usage.
Defaults to 2.0
.
-i INPUT -vf "hwupload, deshake_opencl=smooth_strength=0.5, hwdownload" OUTPUT
-i INPUT -filter_complex "[0:v]format=rgba, hwupload, deshake_opencl=debug=1, hwdownload, format=rgba, format=yuv420p" -v verbose OUTPUT
Apply dilation effect to the video.
This filter replaces the pixel by the local(3x3) maximum.
It accepts the following options:
Limit the maximum change for each plane. Range is [0, 65535]
and default value is 65535
.
If 0
, plane will remain unchanged.
Flag which specifies the pixel to refer to.
Range is [0, 255]
and default value is 255
, i.e. all eight pixels are used.
Flags to local 3x3 coordinates region centered on x
:
1 2 3
4 x 5
6 7 8
-i INPUT -vf "hwupload, dilation_opencl=30:40:50:coordinates=231, hwdownload" OUTPUT
Non-local Means denoise filter through OpenCL, this filter accepts same options as nlmeans.
Overlay one video on top of another.
It takes two inputs and has one output. The first input is the "main" video on which the second input is overlaid. This filter requires same memory layout for all the inputs. So, format conversion may be needed.
The filter accepts the following options:
Set the x coordinate of the overlaid video on the main video.
Default value is 0
.
Set the y coordinate of the overlaid video on the main video.
Default value is 0
.
-i INPUT -i LOGO -filter_complex "[0:v]hwupload[a], [1:v]format=yuv420p, hwupload[b], [a][b]overlay_opencl, hwdownload" OUTPUT
-i INPUT -i LOGO -filter_complex "[0:v]hwupload[a], [1:v]format=yuva420p, hwupload[b], [a][b]overlay_opencl, hwdownload" OUTPUT
Add paddings to the input image, and place the original input at the provided x, y coordinates.
It accepts the following options:
Specify an expression for the size of the output image with the paddings added. If the value for width or height is 0, the corresponding input size is used for the output.
The width expression can reference the value set by the height expression, and vice versa.
The default value of width and height is 0.
Specify the offsets to place the input image at within the padded area, with respect to the top/left border of the output image.
The x expression can reference the value set by the y expression, and vice versa.
The default value of x and y is 0.
If x or y evaluate to a negative number, they’ll be changed so the input image is centered on the padded area.
Specify the color of the padded area. For the syntax of this option, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual.
Pad to an aspect instead to a resolution.
The value for the width, height, x, and y options are expressions containing the following constants:
The input video width and height.
These are the same as in_w and in_h.
The output width and height (the size of the padded area), as specified by the width and height expressions.
These are the same as out_w and out_h.
The x and y offsets as specified by the x and y expressions, or NAN if not yet specified.
same as iw / ih
input sample aspect ratio
input display aspect ratio, it is the same as (iw / ih) * sar
Apply the Prewitt operator (https://en.wikipedia.org/wiki/Prewitt_operator) to input video stream.
The filter accepts the following option:
Set which planes to filter. Default value is 0xf
, by which all planes are processed.
Set value which will be multiplied with filtered result.
Range is [0.0, 65535]
and default value is 1.0
.
Set value which will be added to filtered result.
Range is [-65535, 65535]
and default value is 0.0
.
Filter video using an OpenCL program.
OpenCL program source file.
Kernel name in program.
Number of inputs to the filter. Defaults to 1.
Size of output frames. Defaults to the same as the first input.
The program_opencl
filter also supports the framesync options.
The program source file must contain a kernel function with the given name, which will be run once for each plane of the output. Each run on a plane gets enqueued as a separate 2D global NDRange with one work-item for each pixel to be generated. The global ID offset for each work-item is therefore the coordinates of a pixel in the destination image.
The kernel function needs to take the following arguments:
This image will become the output; the kernel should write all of it.
This is a counter starting from zero and increasing by one for each frame.
These are the most recent images on each input. The kernel may read from them to generate the output, but they can’t be written to.
Example programs:
__kernel void copy(__write_only image2d_t destination, unsigned int index, __read_only image2d_t source) { const sampler_t sampler = CLK_NORMALIZED_COORDS_FALSE; int2 location = (int2)(get_global_id(0), get_global_id(1)); float4 value = read_imagef(source, sampler, location); write_imagef(destination, location, value); }
__kernel void rotate_image(__write_only image2d_t dst, unsigned int index, __read_only image2d_t src) { const sampler_t sampler = (CLK_NORMALIZED_COORDS_FALSE | CLK_FILTER_LINEAR); float angle = (float)index / 100.0f; float2 dst_dim = convert_float2(get_image_dim(dst)); float2 src_dim = convert_float2(get_image_dim(src)); float2 dst_cen = dst_dim / 2.0f; float2 src_cen = src_dim / 2.0f; int2 dst_loc = (int2)(get_global_id(0), get_global_id(1)); float2 dst_pos = convert_float2(dst_loc) - dst_cen; float2 src_pos = { cos(angle) * dst_pos.x - sin(angle) * dst_pos.y, sin(angle) * dst_pos.x + cos(angle) * dst_pos.y }; src_pos = src_pos * src_dim / dst_dim; float2 src_loc = src_pos + src_cen; if (src_loc.x < 0.0f || src_loc.y < 0.0f || src_loc.x > src_dim.x || src_loc.y > src_dim.y) write_imagef(dst, dst_loc, 0.5f); else write_imagef(dst, dst_loc, read_imagef(src, sampler, src_loc)); }
__kernel void blend_images(__write_only image2d_t dst, unsigned int index, __read_only image2d_t src1, __read_only image2d_t src2) { const sampler_t sampler = (CLK_NORMALIZED_COORDS_FALSE | CLK_FILTER_LINEAR); float blend = (cos((float)index / 50.0f) + 1.0f) / 2.0f; int2 dst_loc = (int2)(get_global_id(0), get_global_id(1)); int2 src1_loc = dst_loc * get_image_dim(src1) / get_image_dim(dst); int2 src2_loc = dst_loc * get_image_dim(src2) / get_image_dim(dst); float4 val1 = read_imagef(src1, sampler, src1_loc); float4 val2 = read_imagef(src2, sampler, src2_loc); write_imagef(dst, dst_loc, val1 * blend + val2 * (1.0f - blend)); }
Remap pixels using 2nd: Xmap and 3rd: Ymap input video stream.
Destination pixel at position (X, Y) will be picked from source (x, y) position where x = Xmap(X, Y) and y = Ymap(X, Y). If mapping values are out of range, zero value for pixel will be used for destination pixel.
Xmap and Ymap input video streams must be of same dimensions. Output video stream will have Xmap/Ymap video stream dimensions. Xmap and Ymap input video streams are 32bit float pixel format, single channel.
Specify interpolation used for remapping of pixels.
Allowed values are near
and linear
.
Default value is linear
.
Specify the color of the unmapped pixels. For the syntax of this option,
check the (ffmpeg-utils)"Color" section in the ffmpeg-utils
manual. Default color is black
.
Apply the Roberts cross operator (https://en.wikipedia.org/wiki/Roberts_cross) to input video stream.
The filter accepts the following option:
Set which planes to filter. Default value is 0xf
, by which all planes are processed.
Set value which will be multiplied with filtered result.
Range is [0.0, 65535]
and default value is 1.0
.
Set value which will be added to filtered result.
Range is [-65535, 65535]
and default value is 0.0
.
-i INPUT -vf "hwupload, roberts_opencl=scale=2:delta=10, hwdownload" OUTPUT
Apply the Sobel operator (https://en.wikipedia.org/wiki/Sobel_operator) to input video stream.
The filter accepts the following option:
Set which planes to filter. Default value is 0xf
, by which all planes are processed.
Set value which will be multiplied with filtered result.
Range is [0.0, 65535]
and default value is 1.0
.
Set value which will be added to filtered result.
Range is [-65535, 65535]
and default value is 0.0
.
-i INPUT -vf "hwupload, sobel_opencl=scale=2:delta=10, hwdownload" OUTPUT
Perform HDR(PQ/HLG) to SDR conversion with tone-mapping.
It accepts the following parameters:
Specify the tone-mapping operator to be used. Same as tonemap option in tonemap.
Tune the tone mapping algorithm. same as param option in tonemap.
Apply desaturation for highlights that exceed this level of brightness. The higher the parameter, the more color information will be preserved. This setting helps prevent unnaturally blown-out colors for super-highlights, by (smoothly) turning into white instead. This makes images feel more natural, at the cost of reducing information about out-of-range colors.
The default value is 0.5, and the algorithm here is a little different from the cpu version tonemap currently. A setting of 0.0 disables this option.
The tonemapping algorithm parameters is fine-tuned per each scene. And a threshold is used to detect whether the scene has changed or not. If the distance between the current frame average brightness and the current running average exceeds a threshold value, we would re-calculate scene average and peak brightness. The default value is 0.2.
Specify the output pixel format.
Currently supported formats are:
Set the output color range.
Possible values are:
Default is same as input.
Set the output color primaries.
Possible values are:
Default is same as input.
Set the output transfer characteristics.
Possible values are:
Default is bt709.
Set the output colorspace matrix.
Possible value are:
Default is same as input.
-i INPUT -vf "format=p010,hwupload,tonemap_opencl=t=bt2020:tonemap=linear:format=p010,hwdownload,format=p010" OUTPUT
Sharpen or blur the input video.
It accepts the following parameters:
Set the luma matrix horizontal size.
Range is [1, 23]
and default value is 5
.
Set the luma matrix vertical size.
Range is [1, 23]
and default value is 5
.
Set the luma effect strength.
Range is [-10, 10]
and default value is 1.0
.
Negative values will blur the input video, while positive values will sharpen it, a value of zero will disable the effect.
Set the chroma matrix horizontal size.
Range is [1, 23]
and default value is 5
.
Set the chroma matrix vertical size.
Range is [1, 23]
and default value is 5
.
Set the chroma effect strength.
Range is [-10, 10]
and default value is 0.0
.
Negative values will blur the input video, while positive values will sharpen it, a value of zero will disable the effect.
All parameters are optional and default to the equivalent of the string ’5:5:1.0:5:5:0.0’.
-i INPUT -vf "hwupload, unsharp_opencl=luma_msize_x=7:luma_msize_y=7:luma_amount=2.5, hwdownload" OUTPUT
-i INPUT -vf "hwupload, unsharp_opencl=7:7:-2:7:7:-2, hwdownload" OUTPUT
Cross fade two videos with custom transition effect by using OpenCL.
It accepts the following options:
Set one of possible transition effects.
Select custom transition effect, the actual transition description will be picked from source and kernel options.
Default transition is fade.
OpenCL program source file for custom transition.
Set name of kernel to use for custom transition from program source file.
Set duration of video transition.
Set time of start of transition relative to first video.
The program source file must contain a kernel function with the given name, which will be run once for each plane of the output. Each run on a plane gets enqueued as a separate 2D global NDRange with one work-item for each pixel to be generated. The global ID offset for each work-item is therefore the coordinates of a pixel in the destination image.
The kernel function needs to take the following arguments:
This image will become the output; the kernel should write all of it.
These are the most recent images on each input. The kernel may read from them to generate the output, but they can’t be written to.
Example programs:
__kernel void blend_images(__write_only image2d_t dst, __read_only image2d_t src1, __read_only image2d_t src2, float progress) { const sampler_t sampler = (CLK_NORMALIZED_COORDS_FALSE | CLK_FILTER_LINEAR); int2 p = (int2)(get_global_id(0), get_global_id(1)); float2 rp = (float2)(get_global_id(0), get_global_id(1)); float2 dim = (float2)(get_image_dim(src1).x, get_image_dim(src1).y); rp = rp / dim; float2 dots = (float2)(20.0, 20.0); float2 center = (float2)(0,0); float2 unused; float4 val1 = read_imagef(src1, sampler, p); float4 val2 = read_imagef(src2, sampler, p); bool next = distance(fract(rp * dots, &unused), (float2)(0.5, 0.5)) < (progress / distance(rp, center)); write_imagef(dst, p, next ? val1 : val2); }
VAAPI Video filters are usually used with VAAPI decoder and VAAPI encoder. Below is a description of VAAPI video filters.
To enable compilation of these filters you need to configure FFmpeg with
--enable-vaapi
.
To use vaapi filters, you need to setup the vaapi device correctly. For more information, please read https://trac.ffmpeg.org/wiki/Hardware/VAAPI
Overlay one video on the top of another.
It takes two inputs and has one output. The first input is the "main" video on which the second input is overlaid.
The filter accepts the following options:
Set expressions for the x and y coordinates of the overlaid video on the main video.
Default value is "0" for both expressions.
Set expressions for the width and height the overlaid video on the main video.
Default values are ’overlay_iw’ for ’w’ and ’overlay_ih*w/overlay_iw’ for ’h’.
The expressions can contain the following parameters:
The main input width and height.
The overlay input width and height.
The overlay output width and height.
Position of the overlay layer inside of main
Set transparency of overlaid video. Allowed range is 0.0 to 1.0.
Higher value means lower transparency.
Default value is 1.0
.
See framesync.
See framesync.
See framesync.
This filter also supports the framesync options.
-i INPUT -i LOGO -filter_complex "[0:v]hwupload[a], [1:v]format=yuv420p, hwupload[b], [a][b]overlay_vaapi" OUTPUT
-i INPUT -i LOGO -filter_complex "[0:v]hwupload[a], [1:v]format=yuva420p, hwupload[b], [a][b]overlay_vaapi=x=200:y=100:w=400:h=300:alpha=1.0, hwdownload, format=nv12" OUTPUT
Perform HDR(High Dynamic Range) to SDR(Standard Dynamic Range) conversion with tone-mapping. It maps the dynamic range of HDR10 content to the SDR content. It currently only accepts HDR10 as input.
It accepts the following parameters:
Specify the output pixel format.
Currently supported formats are:
Default is nv12.
Set the output color primaries.
Default is same as input.
Set the output transfer characteristics.
Default is bt709.
Set the output colorspace matrix.
Default is same as input.
tonemap_vaapi=format=p010:t=bt2020-10
Stack input videos horizontally.
This is the VA-API variant of the hstack filter, each input stream may have different height, this filter will scale down/up each input stream while keeping the original aspect.
It accepts the following options:
Stack input videos vertically.
This is the VA-API variant of the vstack filter, each input stream may have different width, this filter will scale down/up each input stream while keeping the original aspect.
It accepts the following options:
Stack video inputs into custom layout.
This is the VA-API variant of the xstack filter, each input stream may have different size, this filter will scale down/up each input stream to the given output size, or the size of the first input stream.
It accepts the following options:
See xstack.
See xstack.
See xstack. Moreover, this permits the user to supply output size for each input stream.
xstack_vaapi=inputs=4:layout=0_0_1920x1080|0_h0_1920x1080|w0_0_1920x1080|w0_h0_1920x1080
See xstack.
Set output size for each input stream when grid is set. If this option is not set, this filter will set output size by default to the size of the first input stream. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
See xstack.
Below is a description of the currently available Vulkan video filters.
To enable compilation of these filters you need to configure FFmpeg with
--enable-vulkan
and either --enable-libglslang
or --enable-libshaderc
.
Running Vulkan filters requires you to initialize a hardware device and to pass that device to all filters in any filter graph.
Initialise a new hardware device of type vulkan called name, using the given device parameters and options in key=value. The following options are supported:
Switches validation layers on if set to 1.
Allocates linear images. Does not apply to decoding.
Disables multiplane images. Does not apply to decoding.
Pass the hardware device called name to all filters in any filter graph.
For more detailed information see https://www.ffmpeg.org/ffmpeg.html#Advanced-Video-options
-init_hw_device vulkan=vk:0 -filter_hw_device vk -i INPUT -vf "hwupload,nlmeans_vulkan,hwdownload" OUTPUT
As Vulkan filters are not able to access frame data in normal memory, all frame data needs to be uploaded (hwupload) to hardware surfaces connected to the appropriate device before being used and then downloaded (hwdownload) back to normal memory. Note that hwupload will upload to a frame with the same layout as the software frame, so it may be necessary to add a format filter immediately before to get the input into the right format and hwdownload does not support all formats on the output - it is usually necessary to insert an additional format filter immediately following in the graph to get the output in a supported format.
Apply an average blur filter, implemented on the GPU using Vulkan.
The filter accepts the following options:
Set horizontal radius size.
Range is [1, 32]
and default value is 3
.
Set vertical radius size. Range is [1, 32]
and default value is 3
.
Set which planes to filter. Default value is 0xf
, by which all planes are processed.
Blend two Vulkan frames into each other.
The blend
filter takes two input streams and outputs one
stream, the first input is the "top" layer and second input is
"bottom" layer. By default, the output terminates when the longest input terminates.
A description of the accepted options follows.
Set blend mode for specific pixel component or all pixel components in case
of all_mode. Default value is normal
.
Available values for component modes are:
Deinterlacer using bwdif, the "Bob Weaver Deinterlacing Filter" algorithm, implemented on the GPU using Vulkan.
It accepts the following parameters:
The interlacing mode to adopt. It accepts one of the following values:
Output one frame for each frame.
Output one frame for each field.
The default value is send_field
.
The picture field parity assumed for the input interlaced video. It accepts one of the following values:
Assume the top field is first.
Assume the bottom field is first.
Enable automatic detection of field parity.
The default value is auto
.
If the interlacing is unknown or the decoder does not export this information,
top field first will be assumed.
Specify which frames to deinterlace. Accepts one of the following values:
Deinterlace all frames.
Only deinterlace frames marked as interlaced.
The default value is all
.
Apply an effect that emulates chromatic aberration. Works best with RGB inputs, but provides a similar effect with YCbCr inputs too.
Horizontal displacement multiplier. Each chroma pixel’s position will be multiplied
by this amount, starting from the center of the image. Default is 0
.
Similarly, this sets the vertical displacement multiplier. Default is 0
.
Video source that creates a Vulkan frame of a solid color. Useful for benchmarking, or overlaying.
It accepts the following parameters:
The color to use. Either a name, or a hexadecimal value.
The default value is black
.
The size of the output frame. Default value is 1920x1080
.
The framerate to output at. Default value is 60
frames per second.
The video duration. Default value is -0.000001
.
The video signal aspect ratio. Default value is 1/1
.
The pixel format of the output Vulkan frames. Default value is yuv444p
.
Set the output YCbCr sample range.
This allows the autodetected value to be overridden as well as allows forcing a specific value used for the output and encoder. If not specified, the range depends on the pixel format. Possible values:
Choose automatically.
Set full range (0-255 in case of 8-bit luma).
Set "MPEG" range (16-235 in case of 8-bit luma).
Flips an image vertically.
Flips an image horizontally.
Flips an image along both the vertical and horizontal axis.
Apply Gaussian blur filter on Vulkan frames.
The filter accepts the following options:
Set horizontal sigma, standard deviation of Gaussian blur. Default is 0.5
.
Set vertical sigma, if negative it will be same as sigma
.
Default is -1
.
Set which planes to filter. By default all planes are filtered.
Set the kernel size along the horizontal axis. Default is 19
.
Set the kernel size along the vertical axis. Default is 0
,
which sets to use the same value as size.
Denoise frames using Non-Local Means algorithm, implemented on the GPU using Vulkan. Supports more pixel formats than nlmeans or nlmeans_opencl, including alpha channel support.
The filter accepts the following options.
Set denoising strength for all components. Default is 1.0. Must be in range [1.0, 100.0].
Set patch size for all planes. Default is 7. Must be odd number in range [0, 99].
Set research size. Default is 15. Must be odd number in range [0, 99].
Set parallelism. Default is 36. Must be a number in the range [1, 168]. Larger values may speed up processing, at the cost of more VRAM. Lower values will slow it down, reducing VRAM usage. Only supported on GPUs with atomic float operations (RDNA3+, Ampere+).
Set denoising strength for a specific component. Default is 1, equal to s. Must be odd number in range [1, 100].
Set patch size for a specific component. Default is 7, equal to p. Must be odd number in range [0, 99].
Overlay one video on top of another.
It takes two inputs and has one output. The first input is the "main" video on which the second input is overlaid. This filter requires all inputs to use the same pixel format. So, format conversion may be needed.
The filter accepts the following options:
Set the x coordinate of the overlaid video on the main video.
Default value is 0
.
Set the y coordinate of the overlaid video on the main video.
Default value is 0
.
Transpose rows with columns in the input video and optionally flip it. For more in depth examples see the transpose video filter, which shares mostly the same options.
It accepts the following parameters:
Specify the transposition direction.
Can assume the following values:
Rotate by 90 degrees counterclockwise and vertically flip. (default)
Rotate by 90 degrees clockwise.
Rotate by 90 degrees counterclockwise.
Rotate by 90 degrees clockwise and vertically flip.
Flip the input video horizontally.
Flip the input video vertically.
Do not apply the transposition if the input geometry matches the one specified by the specified value. It accepts the following values:
Always apply transposition. (default)
Preserve portrait geometry (when height >= width).
Preserve landscape geometry (when width >= height).
Transpose rows with columns in the input video and optionally flip it. For more in depth examples see the transpose video filter, which shares mostly the same options.
It accepts the following parameters:
Specify the transposition direction.
Can assume the following values:
Rotate by 90 degrees counterclockwise and vertically flip. (default)
Rotate by 90 degrees clockwise.
Rotate by 90 degrees counterclockwise.
Rotate by 90 degrees clockwise and vertically flip.
Do not apply the transposition if the input geometry matches the one specified by the specified value. It accepts the following values:
Always apply transposition. (default)
Preserve portrait geometry (when height >= width).
Preserve landscape geometry (when width >= height).
Below is a description of the currently available QSV video filters.
To enable compilation of these filters you need to configure FFmpeg with
--enable-libmfx
or --enable-libvpl
.
To use QSV filters, you need to setup the QSV device correctly. For more information, please read https://trac.ffmpeg.org/wiki/Hardware/QuickSync
Stack input videos horizontally.
This is the QSV variant of the hstack filter, each input stream may have different height, this filter will scale down/up each input stream while keeping the original aspect.
It accepts the following options:
Stack input videos vertically.
This is the QSV variant of the vstack filter, each input stream may have different width, this filter will scale down/up each input stream while keeping the original aspect.
It accepts the following options:
Stack video inputs into custom layout.
This is the QSV variant of the xstack filter.
It accepts the following options:
See xstack.
See xstack.
See xstack. Moreover, this permits the user to supply output size for each input stream.
xstack_qsv=inputs=4:layout=0_0_1920x1080|0_h0_1920x1080|w0_0_1920x1080|w0_h0_1920x1080
See xstack.
Set output size for each input stream when grid is set. If this option is not set, this filter will set output size by default to the size of the first input stream. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
See xstack.
Below is a description of the currently available video sources.
Buffer video frames, and make them available to the filter chain.
This source is mainly intended for a programmatic use, in particular through the interface defined in libavfilter/buffersrc.h.
It accepts the following parameters:
Specify the size (width and height) of the buffered video frames. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
The input video width.
The input video height.
A string representing the pixel format of the buffered video frames. It may be a number corresponding to a pixel format, or a pixel format name.
Specify the timebase assumed by the timestamps of the buffered frames.
Specify the frame rate expected for the video stream.
The sample (pixel) aspect ratio of the input video.
When using a hardware pixel format, this should be a reference to an AVHWFramesContext describing input frames.
For example:
buffer=width=320:height=240:pix_fmt=yuv410p:time_base=1/24:sar=1
will instruct the source to accept video frames with size 320x240 and with format "yuv410p", assuming 1/24 as the timestamps timebase and square pixels (1:1 sample aspect ratio). Since the pixel format with name "yuv410p" corresponds to the number 6 (check the enum AVPixelFormat definition in libavutil/pixfmt.h), this example corresponds to:
buffer=size=320x240:pixfmt=6:time_base=1/24:pixel_aspect=1/1
Alternatively, the options can be specified as a flat string, but this syntax is deprecated:
width:height:pix_fmt:time_base.num:time_base.den:pixel_aspect.num:pixel_aspect.den
Create a pattern generated by an elementary cellular automaton.
The initial state of the cellular automaton can be defined through the filename and pattern options. If such options are not specified an initial state is created randomly.
At each new frame a new row in the video is filled with the result of the cellular automaton next generation. The behavior when the whole frame is filled is defined by the scroll option.
This source accepts the following options:
Read the initial cellular automaton state, i.e. the starting row, from the specified file. In the file, each non-whitespace character is considered an alive cell, a newline will terminate the row, and further characters in the file will be ignored.
Read the initial cellular automaton state, i.e. the starting row, from the specified string.
Each non-whitespace character in the string is considered an alive cell, a newline will terminate the row, and further characters in the string will be ignored.
Set the video rate, that is the number of frames generated per second. Default is 25.
Set the random fill ratio for the initial cellular automaton row. It is a floating point number value ranging from 0 to 1, defaults to 1/PHI.
This option is ignored when a file or a pattern is specified.
Set the seed for filling randomly the initial row, must be an integer included between 0 and UINT32_MAX. If not specified, or if explicitly set to -1, the filter will try to use a good random seed on a best effort basis.
Set the cellular automaton rule, it is a number ranging from 0 to 255. Default value is 110.
Set the size of the output video. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
If filename or pattern is specified, the size is set by default to the width of the specified initial state row, and the height is set to width * PHI.
If size is set, it must contain the width of the specified pattern string, and the specified pattern will be centered in the larger row.
If a filename or a pattern string is not specified, the size value defaults to "320x518" (used for a randomly generated initial state).
If set to 1, scroll the output upward when all the rows in the output have been already filled. If set to 0, the new generated row will be written over the top row just after the bottom row is filled. Defaults to 1.
If set to 1, completely fill the output with generated rows before outputting the first frame. This is the default behavior, for disabling set the value to 0.
If set to 1, stitch the left and right row edges together. This is the default behavior, for disabling set the value to 0.
cellauto=f=pattern:s=200x400
cellauto=ratio=2/3:s=200x200
cellauto=p=@:s=100x400:full=0:rule=18
cellauto=p='@@ @ @@':s=100x400:full=0:rule=18
Video source generated on GPU using Apple’s CoreImage API on OSX.
This video source is a specialized version of the coreimage video filter. Use a core image generator at the beginning of the applied filterchain to generate the content.
The coreimagesrc video source accepts the following options:
List all available generators along with all their respective options as well as possible minimum and maximum values along with the default values.
list_generators=true
Specify the size of the sourced video. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
The default value is 320x240
.
Specify the frame rate of the sourced video, as the number of frames generated per second. It has to be a string in the format frame_rate_num/frame_rate_den, an integer number, a floating point number or a valid video frame rate abbreviation. The default value is "25".
Set the sample aspect ratio of the sourced video.
Set the duration of the sourced video. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax.
If not specified, or the expressed duration is negative, the video is supposed to be generated forever.
Additionally, all options of the coreimage video filter are accepted. A complete filterchain can be used for further processing of the generated input without CPU-HOST transfer. See coreimage documentation and examples for details.
ffmpeg -f lavfi -i coreimagesrc=s=100x100:filter=CIQRCodeGenerator@inputMessage=https\\\\\://FFmpeg.org/@inputCorrectionLevel=H -frames:v 1 QRCode.png
This example is equivalent to the QRCode example of coreimage without the need for a nullsrc video source.
Captures the Windows Desktop via Desktop Duplication API.
The filter exclusively returns D3D11 Hardware Frames, for on-gpu encoding or processing. So an explicit hwdownload is needed for any kind of software processing.
It accepts the following options:
DXGI Output Index to capture.
Usually corresponds to the index Windows has given the screen minus one, so it’s starting at 0.
Defaults to output 0.
Whether to draw the mouse cursor.
Defaults to true.
Only affects hardware cursors. If a game or application renders its own cursor, it’ll always be captured.
Maximum framerate at which the desktop will be captured - the interval between successive frames will not be smaller than the inverse of the framerate. When the desktop is not being updated often enough, the filter will duplicate a previous frame. Note that there is no background buffering going on, so when the filter is not polled often enough then the actual inter-frame interval may be significantly larger.
Defaults to 30 FPS.
Specify the size of the captured video.
Defaults to the full size of the screen.
Cropped from the bottom/right if smaller than screen size.
Horizontal offset of the captured video.
Vertical offset of the captured video.
Desired filter output format. Defaults to 8 Bit BGRA.
It accepts the following values:
Passes all supported output formats to DDA and returns what DDA decides to use.
8 Bit formats always work, and DDA will convert to them if necessary.
Filter initialization will fail if 10 bit format is requested but unavailable.
Capture primary screen and encode using nvenc:
ffmpeg -f lavfi -i ddagrab -c:v h264_nvenc -cq 18 output.mp4
You can also skip the lavfi device and directly use the filter. Also demonstrates downloading the frame and encoding with libx264. Explicit output format specification is required in this case:
ffmpeg -filter_complex ddagrab=output_idx=1:framerate=60,hwdownload,format=bgra -c:v libx264 -crf 18 output.mp4
If you want to capture only a subsection of the desktop, this can be achieved by specifying a smaller size and its offsets into the screen:
ddagrab=video_size=800x600:offset_x=100:offset_y=100
Generate several gradients.
Set frame size. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual. Default value is "640x480".
Set frame rate, expressed as number of frames per second. Default value is "25".
Set 8 colors. Default values for colors is to pick random one.
Set gradient line source and destination points. If negative or out of range, random ones are picked.
Set number of colors to use at once. Allowed range is from 2 to 8. Default value is 2.
Set seed for picking gradient line points.
Set the duration of the sourced video. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax.
If not specified, or the expressed duration is negative, the video is supposed to be generated forever.
Set speed of gradients rotation.
Set type of gradients. Available values are:
Default type is linear.
This source supports the some above options as commands.
Generate a Mandelbrot set fractal, and progressively zoom towards the point specified with start_x and start_y.
This source accepts the following options:
Set the terminal pts value. Default value is 400.
Set the terminal scale value. Must be a floating point value. Default value is 0.3.
Set the inner coloring mode, that is the algorithm used to draw the Mandelbrot fractal internal region.
It shall assume one of the following values:
Set black mode.
Show time until convergence.
Set color based on point closest to the origin of the iterations.
Set period mode.
Default value is mincol.
Set the bailout value. Default value is 10.0.
Set the maximum of iterations performed by the rendering algorithm. Default value is 7189.
Set outer coloring mode. It shall assume one of following values:
Set iteration count mode.
set normalized iteration count mode.
Default value is normalized_iteration_count.
Set frame rate, expressed as number of frames per second. Default value is "25".
Set frame size. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual. Default value is "640x480".
Set the initial scale value. Default value is 3.0.
Set the initial x position. Must be a floating point value between -100 and 100. Default value is -0.743643887037158704752191506114774.
Set the initial y position. Must be a floating point value between -100 and 100. Default value is -0.131825904205311970493132056385139.
Generate various test patterns, as generated by the MPlayer test filter.
The size of the generated video is fixed, and is 256x256. This source is useful in particular for testing encoding features.
This source accepts the following options:
Specify the frame rate of the sourced video, as the number of frames generated per second. It has to be a string in the format frame_rate_num/frame_rate_den, an integer number, a floating point number or a valid video frame rate abbreviation. The default value is "25".
Set the duration of the sourced video. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax.
If not specified, or the expressed duration is negative, the video is supposed to be generated forever.
Set the number or the name of the test to perform. Supported tests are:
Set the maximum number of frames generated for each test, default value is 30.
Default value is "all", which will cycle through the list of all tests.
Some examples:
mptestsrc=t=dc_luma
will generate a "dc_luma" test pattern.
Provide a frei0r source.
To enable compilation of this filter you need to install the frei0r
header and configure FFmpeg with --enable-frei0r
.
This source accepts the following parameters:
The size of the video to generate. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
The framerate of the generated video. It may be a string of the form num/den or a frame rate abbreviation.
The name to the frei0r source to load. For more information regarding frei0r and how to set the parameters, read the frei0r section in the video filters documentation.
A ’|’-separated list of parameters to pass to the frei0r source.
For example, to generate a frei0r partik0l source with size 200x200 and frame rate 10 which is overlaid on the overlay filter main input:
frei0r_src=size=200x200:framerate=10:filter_name=partik0l:filter_params=1234 [overlay]; [in][overlay] overlay
Generate a life pattern.
This source is based on a generalization of John Conway’s life game.
The sourced input represents a life grid, each pixel represents a cell which can be in one of two possible states, alive or dead. Every cell interacts with its eight neighbours, which are the cells that are horizontally, vertically, or diagonally adjacent.
At each interaction the grid evolves according to the adopted rule, which specifies the number of neighbor alive cells which will make a cell stay alive or born. The rule option allows one to specify the rule to adopt.
This source accepts the following options:
Set the file from which to read the initial grid state. In the file, each non-whitespace character is considered an alive cell, and newline is used to delimit the end of each row.
If this option is not specified, the initial grid is generated randomly.
Set the video rate, that is the number of frames generated per second. Default is 25.
Set the random fill ratio for the initial random grid. It is a floating point number value ranging from 0 to 1, defaults to 1/PHI. It is ignored when a file is specified.
Set the seed for filling the initial random grid, must be an integer included between 0 and UINT32_MAX. If not specified, or if explicitly set to -1, the filter will try to use a good random seed on a best effort basis.
Set the life rule.
A rule can be specified with a code of the kind "SNS/BNB", where NS and NB are sequences of numbers in the range 0-8, NS specifies the number of alive neighbor cells which make a live cell stay alive, and NB the number of alive neighbor cells which make a dead cell to become alive (i.e. to "born"). "s" and "b" can be used in place of "S" and "B", respectively.
Alternatively a rule can be specified by an 18-bits integer. The 9
high order bits are used to encode the next cell state if it is alive
for each number of neighbor alive cells, the low order bits specify
the rule for "borning" new cells. Higher order bits encode for an
higher number of neighbor cells.
For example the number 6153 = (12<<9)+9
specifies a stay alive
rule of 12 and a born rule of 9, which corresponds to "S23/B03".
Default value is "S23/B3", which is the original Conway’s game of life rule, and will keep a cell alive if it has 2 or 3 neighbor alive cells, and will born a new cell if there are three alive cells around a dead cell.
Set the size of the output video. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
If filename is specified, the size is set by default to the same size of the input file. If size is set, it must contain the size specified in the input file, and the initial grid defined in that file is centered in the larger resulting area.
If a filename is not specified, the size value defaults to "320x240" (used for a randomly generated initial grid).
If set to 1, stitch the left and right grid edges together, and the top and bottom edges also. Defaults to 1.
Set cell mold speed. If set, a dead cell will go from death_color to mold_color with a step of mold. mold can have a value from 0 to 255.
Set the color of living (or new born) cells.
Set the color of dead cells. If mold is set, this is the first color used to represent a dead cell.
Set mold color, for definitely dead and moldy cells.
For the syntax of these 3 color options, check the (ffmpeg-utils)"Color" section in the ffmpeg-utils manual.
life=f=pattern:s=300x300
life=ratio=2/3:s=200x200
life=rule=S14/B34
ffplay
:
ffplay -f lavfi life=s=300x200:mold=10:r=60:ratio=0.1:death_color=#C83232:life_color=#00ff00,scale=1200:800:flags=16
The allrgb
source returns frames of size 4096x4096 of all rgb colors.
The allyuv
source returns frames of size 4096x4096 of all yuv colors.
The color
source provides an uniformly colored input.
The colorchart
source provides a colors checker chart.
The colorspectrum
source provides a color spectrum input.
The haldclutsrc
source provides an identity Hald CLUT. See also
haldclut filter.
The nullsrc
source returns unprocessed video frames. It is
mainly useful to be employed in analysis / debugging tools, or as the
source for filters which ignore the input data.
The pal75bars
source generates a color bars pattern, based on
EBU PAL recommendations with 75% color levels.
The pal100bars
source generates a color bars pattern, based on
EBU PAL recommendations with 100% color levels.
The rgbtestsrc
source generates an RGB test pattern useful for
detecting RGB vs BGR issues. You should see a red, green and blue
stripe from top to bottom.
The smptebars
source generates a color bars pattern, based on
the SMPTE Engineering Guideline EG 1-1990.
The smptehdbars
source generates a color bars pattern, based on
the SMPTE RP 219-2002.
The testsrc
source generates a test video pattern, showing a
color pattern, a scrolling gradient and a timestamp. This is mainly
intended for testing purposes.
The testsrc2
source is similar to testsrc, but supports more
pixel formats instead of just rgb24
. This allows using it as an
input for other tests without requiring a format conversion.
The yuvtestsrc
source generates an YUV test pattern. You should
see a y, cb and cr stripe from top to bottom.
The sources accept the following parameters:
Specify the level of the Hald CLUT, only available in the haldclutsrc
source. A level of N
generates a picture of N*N*N
by N*N*N
pixels to be used as identity matrix for 3D lookup tables. Each component is
coded on a 1/(N*N)
scale.
Specify the color of the source, only available in the color
source. For the syntax of this option, check the
(ffmpeg-utils)"Color" section in the ffmpeg-utils manual.
Specify the size of the sourced video. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
The default value is 320x240
.
This option is not available with the allrgb
, allyuv
, and
haldclutsrc
filters.
Specify the frame rate of the sourced video, as the number of frames generated per second. It has to be a string in the format frame_rate_num/frame_rate_den, an integer number, a floating point number or a valid video frame rate abbreviation. The default value is "25".
Set the duration of the sourced video. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax.
If not specified, or the expressed duration is negative, the video is supposed to be generated forever.
Since the frame rate is used as time base, all frames including the last one will have their full duration. If the specified duration is not a multiple of the frame duration, it will be rounded up.
Set the sample aspect ratio of the sourced video.
Specify the alpha (opacity) of the background, only available in the
testsrc2
source. The value must be between 0 (fully transparent) and
255 (fully opaque, the default).
Set the number of decimals to show in the timestamp, only available in the
testsrc
source.
The displayed timestamp value will correspond to the original timestamp value multiplied by the power of 10 of the specified value. Default value is 0.
Set the type of the color spectrum, only available in the
colorspectrum
source. Can be one of the following:
Set patch size of single color patch, only available in the
colorchart
source. Default is 64x64
.
Set colorchecker colors preset, only available in the
colorchart
source.
Available values are:
Default value is reference
.
testsrc=duration=5.3:size=qcif:rate=10
color=c=red@0.2:s=qcif:r=10
nullsrc
can be used. The
following command generates noise in the luma plane by employing
the geq
filter:
nullsrc=s=256x256, geq=random(1)*255:128:128
The color
source supports the following commands:
Set the color of the created image. Accepts the same syntax of the corresponding color option.
Generate video using an OpenCL program.
OpenCL program source file.
Kernel name in program.
Size of frames to generate. This must be set.
Pixel format to use for the generated frames. This must be set.
Number of frames generated every second. Default value is ’25’.
For details of how the program loading works, see the program_opencl filter.
Example programs:
__kernel void ramp(__write_only image2d_t dst, unsigned int index) { int2 loc = (int2)(get_global_id(0), get_global_id(1)); float4 val; val.xy = val.zw = convert_float2(loc) / convert_float2(get_image_dim(dst)); write_imagef(dst, loc, val); }
__kernel void sierpinski_carpet(__write_only image2d_t dst, unsigned int index) { int2 loc = (int2)(get_global_id(0), get_global_id(1)); float4 value = 0.0f; int x = loc.x + index; int y = loc.y + index; while (x > 0 || y > 0) { if (x % 3 == 1 && y % 3 == 1) { value = 1.0f; break; } x /= 3; y /= 3; } write_imagef(dst, loc, value); }
Generate a Sierpinski carpet/triangle fractal, and randomly pan around.
This source accepts the following options:
Set frame size. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual. Default value is "640x480".
Set frame rate, expressed as number of frames per second. Default value is "25".
Set seed which is used for random panning.
Set max jump for single pan destination. Allowed range is from 1 to 10000.
Set fractal type, can be default carpet
or triangle
.
Generate a zoneplate test video pattern.
This source accepts the following options:
Set frame size. For the syntax of this option, check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual. Default value is "320x240".
Set frame rate, expressed as number of frames per second. Default value is "25".
Set the duration of the sourced video. See (ffmpeg-utils)the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax.
If not specified, or the expressed duration is negative, the video is supposed to be generated forever.
Set the sample aspect ratio of the sourced video.
Set precision in bits for look-up table for sine calculations. Default value is 10. Allowed range is from 4 to 16.
Set horizontal axis offset for output signal. Default value is 0.
Set vertical axis offset for output signal. Default value is 0.
Set time axis offset for output signal. Default value is 0.
Set 0-order, constant added to signal phase. Default value is 0.
Set 1-order, phase factor multiplier for horizontal axis. Default value is 0.
Set 1-order, phase factor multiplier for vertical axis. Default value is 0.
Set 1-order, phase factor multiplier for time axis. Default value is 0.
Set phase factor multipliers for combination of spatial and temporal axis. Default value is 0.
Set 2-order, phase factor multiplier for horizontal axis. Default value is 0.
Set 2-order, phase factor multiplier for vertical axis. Default value is 0.
Set 2-order, phase factor multiplier for time axis. Default value is 0.
Set the constant added to final phase to produce chroma-blue component of signal. Default value is 0.
Set the constant added to final phase to produce chroma-red component of signal. Default value is 0.
This source supports the some above options as commands.
zoneplate=ku=512:kv=0:kt2=0:kx2=256:s=wvga:xo=-426:kt=11
zoneplate=ku=512:kv=0:kt2=0:ky2=156:s=wvga:yo=-240:kt=11
zoneplate=ku=512:kv=100:kt2=0:ky2=256:kx2=556:s=wvga:yo=0:kt=11
Below is a description of the currently available video sinks.
Buffer video frames, and make them available to the end of the filter graph.
This sink is mainly intended for programmatic use, in particular through the interface defined in libavfilter/buffersink.h or the options system.
It accepts a pointer to an AVBufferSinkContext structure, which
defines the incoming buffers’ formats, to be passed as the opaque
parameter to avfilter_init_filter
for initialization.
Null video sink: do absolutely nothing with the input video. It is mainly useful as a template and for use in analysis / debugging tools.
Below is a description of the currently available multimedia filters.
Convert input audio to 3d scope video output.
The filter accepts the following options:
Set frame rate, expressed as number of frames per second. Default value is "25".
Specify the video size for the output. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is hd720
.
Set the camera field of view. Default is 90 degrees. Allowed range is from 40 to 150.
Set the camera roll.
Set the camera pitch.
Set the camera yaw.
Set the camera zoom on X-axis.
Set the camera zoom on Y-axis.
Set the camera zoom on Z-axis.
Set the camera position on X-axis.
Set the camera position on Y-axis.
Set the camera position on Z-axis.
Set the length of displayed audio waves in number of frames.
Filter supports the some above options as commands.
Convert input audio to a video output, displaying the audio bit scope.
The filter accepts the following options:
Set frame rate, expressed as number of frames per second. Default value is "25".
Specify the video size for the output. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is 1024x256
.
Specify list of colors separated by space or by ’|’ which will be used to draw channels. Unrecognized or missing colors will be replaced by white color.
Set output mode. Can be bars
or trace
. Default is bars
.
See graphmonitor.
Convert input audio to a video output, displaying the volume histogram.
The filter accepts the following options:
Specify how histogram is calculated.
It accepts the following values:
Use single histogram for all channels.
Use separate histogram for each channel.
Default is single
.
Set frame rate, expressed as number of frames per second. Default value is "25".
Specify the video size for the output. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is hd720
.
Set display scale.
It accepts the following values:
logarithmic
square root
cubic root
linear
reverse logarithmic
Default is log
.
Set amplitude scale.
It accepts the following values:
logarithmic
linear
Default is log
.
Set how much frames to accumulate in histogram. Default is 1. Setting this to -1 accumulates all frames.
Set histogram ratio of window height.
Set sonogram sliding.
It accepts the following values:
replace old rows with new ones.
scroll from top to bottom.
Default is replace
.
Set histogram mode.
It accepts the following values:
Use absolute values of samples.
Use untouched values of samples.
Default is abs
.
Measures phase of input audio, which is exported as metadata lavfi.aphasemeter.phase
,
representing mean phase of current audio frame. A video output can also be produced and is
enabled by default. The audio is passed through as first output.
Audio will be rematrixed to stereo if it has a different channel layout. Phase value is in
range [-1, 1]
where -1
means left and right channels are completely out of phase
and 1
means channels are in phase.
The filter accepts the following options, all related to its video output:
Set the output frame rate. Default value is 25
.
Set the video size for the output. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is 800x400
.
Specify the red, green, blue contrast. Default values are 2
,
7
and 1
.
Allowed range is [0, 255]
.
Set color which will be used for drawing median phase. If color is
none
which is default, no median phase value will be drawn.
Enable video output. Default is enabled.
The filter also detects out of phase and mono sequences in stereo streams. It logs the sequence start, end and duration when it lasts longer or as long as the minimum set.
The filter accepts the following options for this detection:
Enable mono and out of phase detection. Default is disabled.
Set phase tolerance for mono detection, in amplitude ratio. Default is 0
.
Allowed range is [0, 1]
.
Set angle threshold for out of phase detection, in degree. Default is 170
.
Allowed range is [90, 180]
.
Set mono or out of phase duration until notification, expressed in seconds. Default is 2
.
ffmpeg
to detect 1 second of mono with 0.001 phase tolerance:
ffmpeg -i stereo.wav -af aphasemeter=video=0:phasing=1:duration=1:tolerance=0.001 -f null -
Convert input audio to a video output, representing the audio vector scope.
The filter is used to measure the difference between channels of stereo audio stream. A monaural signal, consisting of identical left and right signal, results in straight vertical line. Any stereo separation is visible as a deviation from this line, creating a Lissajous figure. If the straight (or deviation from it) but horizontal line appears this indicates that the left and right channels are out of phase.
The filter accepts the following options:
Set the vectorscope mode.
Available values are:
Lissajous rotated by 45 degrees.
Same as above but not rotated.
Shape resembling half of circle.
Default value is ‘lissajous’.
Set the video size for the output. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is 400x400
.
Set the output frame rate. Default value is 25
.
Specify the red, green, blue and alpha contrast. Default values are 40
,
160
, 80
and 255
.
Allowed range is [0, 255]
.
Specify the red, green, blue and alpha fade. Default values are 15
,
10
, 5
and 5
.
Allowed range is [0, 255]
.
Set the zoom factor. Default value is 1
. Allowed range is [0, 10]
.
Values lower than 1 will auto adjust zoom factor to maximal possible value.
Set the vectorscope drawing mode.
Available values are:
Draw dot for each sample.
Draw line between previous and current sample.
Draw anti-aliased line between previous and current sample.
Default value is ‘dot’.
Specify amplitude scale of audio samples.
Available values are:
Linear.
Square root.
Cubic root.
Logarithmic.
Swap left channel axis with right channel axis.
Mirror axis.
No mirror.
Mirror only x axis.
Mirror only y axis.
Mirror both axis.
ffplay
:
ffplay -f lavfi 'amovie=input.mp3, asplit [a][out1]; [a] avectorscope=zoom=1.3:rc=2:gc=200:bc=10:rf=1:gf=8:bf=7 [out0]'
This filter supports the all above options as commands except options size
and rate
.
Benchmark part of a filtergraph.
The filter accepts the following options:
Start or stop a timer.
Available values are:
Get the current time, set it as frame metadata (using the key
lavfi.bench.start_time
), and forward the frame to the next filter.
Get the current time and fetch the lavfi.bench.start_time
metadata from
the input frame metadata to get the time difference. Time difference, average,
maximum and minimum time (respectively t
, avg
, max
and
min
) are then printed. The timestamps are expressed in seconds.
bench=start,selectivecolor=reds=-.2 .12 -.49,bench=stop
Concatenate audio and video streams, joining them together one after the other.
The filter works on segments of synchronized video and audio streams. All segments must have the same number of streams of each type, and that will also be the number of streams at output.
The filter accepts the following options:
Set the number of segments. Default is 2.
Set the number of output video streams, that is also the number of video streams in each segment. Default is 1.
Set the number of output audio streams, that is also the number of audio streams in each segment. Default is 0.
Activate unsafe mode: do not fail if segments have a different format.
The filter has v+a outputs: first v video outputs, then a audio outputs.
There are nx(v+a) inputs: first the inputs for the first segment, in the same order as the outputs, then the inputs for the second segment, etc.
Related streams do not always have exactly the same duration, for various reasons including codec frame size or sloppy authoring. For that reason, related synchronized streams (e.g. a video and its audio track) should be concatenated at once. The concat filter will use the duration of the longest stream in each segment (except the last one), and if necessary pad shorter audio streams with silence.
For this filter to work correctly, all segments must start at timestamp 0.
All corresponding streams must have the same parameters in all segments; the filtering system will automatically select a common pixel format for video streams, and a common sample format, sample rate and channel layout for audio streams, but other settings, such as resolution, must be converted explicitly by the user.
Different frame rates are acceptable but will result in variable frame rate at output; be sure to configure the output file to handle it.
ffmpeg -i opening.mkv -i episode.mkv -i ending.mkv -filter_complex \ '[0:0] [0:1] [0:2] [1:0] [1:1] [1:2] [2:0] [2:1] [2:2] concat=n=3:v=1:a=2 [v] [a1] [a2]' \ -map '[v]' -map '[a1]' -map '[a2]' output.mkv
movie=part1.mp4, scale=512:288 [v1] ; amovie=part1.mp4 [a1] ; movie=part2.mp4, scale=512:288 [v2] ; amovie=part2.mp4 [a2] ; [v1] [v2] concat [outv] ; [a1] [a2] concat=v=0:a=1 [outa]
Note that a desync will happen at the stitch if the audio and video streams do not have exactly the same duration in the first file.
EBU R128 scanner filter. This filter takes an audio stream and analyzes its loudness
level. By default, it logs a message at a frequency of 10Hz with the
Momentary loudness (identified by M
), Short-term loudness (S
),
Integrated loudness (I
) and Loudness Range (LRA
).
The filter can only analyze streams which have sample format is double-precision floating point. The input stream will be converted to this specification, if needed. Users may need to insert aformat and/or aresample filters after this filter to obtain the original parameters.
The filter also has a video output (see the video option) with a real time graph to observe the loudness evolution. The graphic contains the logged message mentioned above, so it is not printed anymore when this option is set, unless the verbose logging is set. The main graphing area contains the short-term loudness (3 seconds of analysis), and the gauge on the right is for the momentary loudness (400 milliseconds), but can optionally be configured to instead display short-term loudness (see gauge).
The green area marks a +/- 1LU target range around the target loudness (-23LUFS by default, unless modified through target).
More information about the Loudness Recommendation EBU R128 on http://tech.ebu.ch/loudness.
The filter accepts the following options:
Activate the video output. The audio stream is passed unchanged whether this
option is set or no. The video stream will be the first output stream if
activated. Default is 0
.
Set the video size. This option is for video only. For the syntax of this
option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default and minimum resolution is 640x480
.
Set the EBU scale meter. Default is 9
. Common values are 9
and
18
, respectively for EBU scale meter +9 and EBU scale meter +18. Any
other integer value between this range is allowed.
Set metadata injection. If set to 1
, the audio input will be segmented
into 100ms output frames, each of them containing various loudness information
in metadata. All the metadata keys are prefixed with lavfi.r128.
.
Default is 0
.
Force the frame logging level.
Available values are:
logging disabled
information logging level
verbose logging level
By default, the logging level is set to info. If the video or the metadata options are set, it switches to verbose.
Set peak mode(s).
Available modes can be cumulated (the option is a flag
type). Possible
values are:
Disable any peak mode (default).
Enable sample-peak mode.
Simple peak mode looking for the higher sample value. It logs a message
for sample-peak (identified by SPK
).
Enable true-peak mode.
If enabled, the peak lookup is done on an over-sampled version of the input
stream for better peak accuracy. It logs a message for true-peak.
(identified by TPK
) and true-peak per frame (identified by FTPK
).
This mode requires a build with libswresample
.
Treat mono input files as "dual mono". If a mono file is intended for playback
on a stereo system, its EBU R128 measurement will be perceptually incorrect.
If set to true
, this option will compensate for this effect.
Multi-channel input files are not affected by this option.
Set a specific pan law to be used for the measurement of dual mono files. This parameter is optional, and has a default value of -3.01dB.
Set a specific target level (in LUFS) used as relative zero in the visualization. This parameter is optional and has a default value of -23LUFS as specified by EBU R128. However, material published online may prefer a level of -16LUFS (e.g. for use with podcasts or video platforms).
Set the value displayed by the gauge. Valid values are momentary
and s
shortterm
. By default the momentary value will be used, but in certain
scenarios it may be more useful to observe the short term value instead (e.g.
live mixing).
Sets the display scale for the loudness. Valid parameters are absolute
(in LUFS) or relative
(LU) relative to the target. This only affects the
video output, not the summary or continuous log output.
Read-only exported value for measured integrated loudness, in LUFS.
Read-only exported value for measured loudness range, in LU.
Read-only exported value for measured LRA low, in LUFS.
Read-only exported value for measured LRA high, in LUFS.
Read-only exported value for measured sample peak, in dBFS.
Read-only exported value for measured true peak, in dBFS.
ffplay
, with a EBU scale meter +18:
ffplay -f lavfi -i "amovie=input.mp3,ebur128=video=1:meter=18 [out0][out1]"
ffmpeg
:
ffmpeg -nostats -i input.mp3 -filter_complex ebur128 -f null -
Temporally interleave frames from several inputs.
interleave
works with video inputs, ainterleave
with audio.
These filters read frames from several inputs and send the oldest queued frame to the output.
Input streams must have well defined, monotonically increasing frame timestamp values.
In order to submit one frame to output, these filters need to enqueue at least one frame for each input, so they cannot work in case one input is not yet terminated and will not receive incoming frames.
For example consider the case when one input is a select
filter
which always drops input frames. The interleave
filter will keep
reading from that input, but it will never be able to send new frames
to output until the input sends an end-of-stream signal.
Also, depending on inputs synchronization, the filters will drop frames in case one input receives more frames than the other ones, and the queue is already filled.
These filters accept the following options:
Set the number of different inputs, it is 2 by default.
How to determine the end-of-stream.
The duration of the longest input. (default)
The duration of the shortest input.
The duration of the first input.
ffmpeg
:
ffmpeg -i bambi.avi -i pr0n.mkv -filter_complex "[0:v][1:v] interleave" out.avi
select='if(gt(random(0), 0.2), 1, 2)':n=2 [tmp], boxblur=2:2, [tmp] interleave
Measure filtering latency.
Report previous filter filtering latency, delay in number of audio samples for audio filters or number of video frames for video filters.
On end of input stream, filter will report min and max measured latency for previous running filter in filtergraph.
Manipulate frame metadata.
This filter accepts the following options:
Set mode of operation of the filter.
Can be one of the following:
If both value
and key
is set, select frames
which have such metadata. If only key
is set, select
every frame that has such key in metadata.
Add new metadata key
and value
. If key is already available
do nothing.
Modify value of already present key.
If value
is set, delete only keys that have such value.
Otherwise, delete key. If key
is not set, delete all metadata values in
the frame.
Print key and its value if metadata was found. If key
is not set print all
metadata values available in frame.
Set key used with all modes. Must be set for all modes except print
and delete
.
Set metadata value which will be used. This option is mandatory for
modify
and add
mode.
Which function to use when comparing metadata value and value
.
Can be one of following:
Values are interpreted as strings, returns true if metadata value is same as value
.
Values are interpreted as strings, returns true if metadata value starts with
the value
option string.
Values are interpreted as floats, returns true if metadata value is less than value
.
Values are interpreted as floats, returns true if value
is equal with metadata value.
Values are interpreted as floats, returns true if metadata value is greater than value
.
Values are interpreted as floats, returns true if expression from option expr
evaluates to true.
Values are interpreted as strings, returns true if metadata value ends with
the value
option string.
Set expression which is used when function
is set to expr
.
The expression is evaluated through the eval API and can contain the following
constants:
Float representation of value
from metadata key.
Float representation of value
as supplied by user in value
option.
If specified in print
mode, output is written to the named file. Instead of
plain filename any writable url can be specified. Filename “-” is a shorthand
for standard output. If file
option is not set, output is written to the log
with AV_LOG_INFO loglevel.
Reduces buffering in print mode when output is written to a URL set using file.
lavfi.signalstats.YDIF
with values
between 0 and 1.
signalstats,metadata=print:key=lavfi.signalstats.YDIF:value=0:function=expr:expr='between(VALUE1,0,1)'
silencedetect,ametadata=mode=print:file=metadata.txt
metadata=mode=print:file='pipe\:4'
Set read/write permissions for the output frames.
These filters are mainly aimed at developers to test direct path in the following filter in the filtergraph.
The filters accept the following options:
Select the permissions mode.
It accepts the following values:
Do nothing. This is the default.
Set all the output frames read-only.
Set all the output frames directly writable.
Make the frame read-only if writable, and writable if read-only.
Set each output frame read-only or writable randomly.
Set the seed for the random mode, must be an integer included between
0
and UINT32_MAX
. If not specified, or if explicitly set to
-1
, the filter will try to use a good random seed on a best effort
basis.
Note: in case of auto-inserted filter between the permission filter and the following one, the permission might not be received as expected in that following filter. Inserting a format or aformat filter before the perms/aperms filter can avoid this problem.
Slow down filtering to match real time approximately.
These filters will pause the filtering for a variable amount of time to
match the output rate with the input timestamps.
They are similar to the re option to ffmpeg
.
They accept the following options:
Time limit for the pauses. Any pause longer than that will be considered a timestamp discontinuity and reset the timer. Default is 2 seconds.
Speed factor for processing. The value must be a float larger than zero. Values larger than 1.0 will result in faster than realtime processing, smaller will slow processing down. The limit is automatically adapted accordingly. Default is 1.0.
A processing speed faster than what is possible without these filters cannot be achieved.
Both filters supports the all above options as commands.
Split single input stream into multiple streams.
This filter does opposite of concat filters.
segment
works on video frames, asegment
on audio samples.
This filter accepts the following options:
Timestamps of output segments separated by ’|’. The first segment will run from the beginning of the input stream. The last segment will run until the end of the input stream
Exact frame/sample count to split the segments.
In all cases, prefixing an each segment with ’+’ will make it relative to the previous segment.
asegment=timestamps="60|150"
Select frames to pass in output.
This filter accepts the following options:
Set expression, which is evaluated for each input frame.
If the expression is evaluated to zero, the frame is discarded.
If the evaluation result is negative or NaN, the frame is sent to the
first output; otherwise it is sent to the output with index
ceil(val)-1
, assuming that the input index starts from 0.
For example a value of 1.2
corresponds to the output with index
ceil(1.2)-1 = 2-1 = 1
, that is the second output.
Set the number of outputs. The output to which to send the selected frame is based on the result of the evaluation. Default value is 1.
The expression can contain the following constants:
The (sequential) number of the filtered frame, starting from 0.
The (sequential) number of the selected frame, starting from 0.
The sequential number of the last selected frame. It’s NAN if undefined.
The timebase of the input timestamps.
The PTS (Presentation TimeStamp) of the filtered frame, expressed in TB units. It’s NAN if undefined.
The PTS of the filtered frame, expressed in seconds. It’s NAN if undefined.
The PTS of the previously filtered frame. It’s NAN if undefined.
The PTS of the last previously filtered frame. It’s NAN if undefined.
The PTS of the last previously selected frame, expressed in seconds. It’s NAN if undefined.
The first PTS in the stream which is not NAN. It remains NAN if not found.
The first PTS, in seconds, in the stream which is not NAN. It remains NAN if not found.
The type of the filtered frame. It can assume one of the following values:
The frame interlace type. It can assume one of the following values:
The frame is progressive (not interlaced).
The frame is top-field-first.
The frame is bottom-field-first.
the number of selected samples before the current frame
the number of samples in the current frame
the input sample rate
This is 1 if the filtered frame is a key-frame, 0 otherwise.
the position in the file of the filtered frame, -1 if the information is not available (e.g. for synthetic video); deprecated, do not use
value between 0 and 1 to indicate a new scene; a low value reflects a low probability for the current frame to introduce a new scene, while a higher value means the current frame is more likely to be one (see the example below)
The concat demuxer can select only part of a concat input file by setting an inpoint and an outpoint, but the output packets may not be entirely contained in the selected interval. By using this variable, it is possible to skip frames generated by the concat demuxer which are not exactly contained in the selected interval.
This works by comparing the frame pts against the lavf.concat.start_time and the lavf.concat.duration packet metadata values which are also present in the decoded frames.
The concatdec_select variable is -1 if the frame pts is at least start_time and either the duration metadata is missing or the frame pts is less than start_time + duration, 0 otherwise, and NaN if the start_time metadata is missing.
That basically means that an input frame is selected if its pts is within the interval set by the concat demuxer.
The default value of the select expression is "1".
select
The example above is the same as:
select=1
select=0
select='eq(pict_type\,I)'
select='not(mod(n\,100))'
select=between(t\,10\,20)
select=between(t\,10\,20)*eq(pict_type\,I)
select='isnan(prev_selected_t)+gte(t-prev_selected_t\,10)'
aselect='gt(samples_n\,100)'
ffmpeg -i video.avi -vf select='gt(scene\,0.4)',scale=160:120,tile -frames:v 1 preview.png
Comparing scene against a value between 0.3 and 0.5 is generally a sane choice.
select=n=2:e='mod(n, 2)+1' [odd][even]; [odd] pad=h=2*ih [tmp]; [tmp][even] overlay=y=h
ffmpeg -copyts -vsync 0 -segment_time_metadata 1 -i input.ffconcat -vf select=concatdec_select -af aselect=concatdec_select output.avi
Send commands to filters in the filtergraph.
These filters read commands to be sent to other filters in the filtergraph.
sendcmd
must be inserted between two video filters,
asendcmd
must be inserted between two audio filters, but apart
from that they act the same way.
The specification of commands can be provided in the filter arguments with the commands option, or in a file specified by the filename option.
These filters accept the following options:
Set the commands to be read and sent to the other filters.
Set the filename of the commands to be read and sent to the other filters.
A commands description consists of a sequence of interval specifications, comprising a list of commands to be executed when a particular event related to that interval occurs. The occurring event is typically the current frame time entering or leaving a given time interval.
An interval is specified by the following syntax:
START[-END] COMMANDS;
The time interval is specified by the START and END times. END is optional and defaults to the maximum time.
The current frame time is considered within the specified interval if it is included in the interval [START, END), that is when the time is greater or equal to START and is lesser than END.
COMMANDS consists of a sequence of one or more command specifications, separated by ",", relating to that interval. The syntax of a command specification is given by:
[FLAGS] TARGET COMMAND ARG
FLAGS is optional and specifies the type of events relating to the time interval which enable sending the specified command, and must be a non-null sequence of identifier flags separated by "+" or "|" and enclosed between "[" and "]".
The following flags are recognized:
The command is sent when the current frame timestamp enters the specified interval. In other words, the command is sent when the previous frame timestamp was not in the given interval, and the current is.
The command is sent when the current frame timestamp leaves the specified interval. In other words, the command is sent when the previous frame timestamp was in the given interval, and the current is not.
The command ARG is interpreted as expression and result of expression is passed as ARG.
The expression is evaluated through the eval API and can contain the following constants:
Original position in the file of the frame, or undefined if undefined for the current frame. Deprecated, do not use.
The presentation timestamp in input.
The count of the input frame for video or audio, starting from 0.
The time in seconds of the current frame.
The start time in seconds of the current command interval.
The end time in seconds of the current command interval.
The interpolated time of the current command interval, TI = (T - TS) / (TE - TS).
The video frame width.
The video frame height.
If FLAGS is not specified, a default value of [enter]
is
assumed.
TARGET specifies the target of the command, usually the name of the filter class or a specific filter instance name.
COMMAND specifies the name of the command for the target filter.
ARG is optional and specifies the optional list of argument for the given COMMAND.
Between one interval specification and another, whitespaces, or
sequences of characters starting with #
until the end of line,
are ignored and can be used to annotate comments.
A simplified BNF description of the commands specification syntax follows:
COMMAND_FLAG ::= "enter" | "leave" COMMAND_FLAGS ::= COMMAND_FLAG [(+|"|")COMMAND_FLAG] COMMAND ::= ["[" COMMAND_FLAGS "]"] TARGET COMMAND [ARG] COMMANDS ::= COMMAND [,COMMANDS] INTERVAL ::= START[-END] COMMANDS INTERVALS ::= INTERVAL[;INTERVALS]
asendcmd=c='4.0 atempo tempo 1.5',atempo
asendcmd=c='4.0 atempo@my tempo 1.5',atempo@my
# show text in the interval 5-10 5.0-10.0 [enter] drawtext reinit 'fontfile=FreeSerif.ttf:text=hello world', [leave] drawtext reinit 'fontfile=FreeSerif.ttf:text='; # desaturate the image in the interval 15-20 15.0-20.0 [enter] hue s 0, [enter] drawtext reinit 'fontfile=FreeSerif.ttf:text=nocolor', [leave] hue s 1, [leave] drawtext reinit 'fontfile=FreeSerif.ttf:text=color'; # apply an exponential saturation fade-out effect, starting from time 25 25 [enter] hue s exp(25-t)
A filtergraph allowing to read and process the above command list stored in a file test.cmd, can be specified with:
sendcmd=f=test.cmd,drawtext=fontfile=FreeSerif.ttf:text='',hue
Change the PTS (presentation timestamp) of the input frames.
setpts
works on video frames, asetpts
on audio frames.
This filter accepts the following options:
The expression which is evaluated for each frame to construct its timestamp.
The expression is evaluated through the eval API and can contain the following constants:
frame rate, only defined for constant frame-rate video
The presentation timestamp in input
The count of the input frame for video or the number of consumed samples, not including the current frame for audio, starting from 0.
The number of consumed samples, not including the current frame (only audio)
The number of samples in the current frame (only audio)
The audio sample rate.
The PTS of the first frame.
the time in seconds of the first frame
State whether the current frame is interlaced.
the time in seconds of the current frame
original position in the file of the frame, or undefined if undefined for the current frame; deprecated, do not use
The previous input PTS.
previous input time in seconds
The previous output PTS.
previous output time in seconds
The wallclock (RTC) time in microseconds. This is deprecated, use time(0) instead.
The wallclock (RTC) time at the start of the movie in microseconds.
The timebase of the input timestamps.
Time of the first frame after command was applied or time of the first frame if no commands.
setpts=PTS-STARTPTS
setpts=0.5*PTS
setpts=2.0*PTS
setpts=N/(25*TB)
setpts='1/(25*TB) * (N + 0.05 * sin(N*2*PI/25))'
setpts=PTS+10/TB
setpts='(RTCTIME - RTCSTART) / (TB * 1000000)'
asetpts=N/SR/TB
Both filters support all above options as commands.
Force color range for the output video frame.
The setrange
filter marks the color range property for the
output frames. It does not change the input frame, but only sets the
corresponding property, which affects how the frame is treated by
following filters.
The filter accepts the following options:
Available values are:
Keep the same color range property.
Set the color range as unspecified.
Set the color range as limited.
Set the color range as full.
Set the timebase to use for the output frames timestamps. It is mainly useful for testing timebase configuration.
It accepts the following parameters:
The expression which is evaluated into the output timebase.
The value for tb is an arithmetic expression representing a rational. The expression can contain the constants "AVTB" (the default timebase), "intb" (the input timebase) and "sr" (the sample rate, audio only). Default value is "intb".
settb=expr=1/25
settb=expr=0.1
settb=1+0.001
settb=2*intb
settb=AVTB
Convert input audio to a video output representing frequency spectrum logarithmically using Brown-Puckette constant Q transform algorithm with direct frequency domain coefficient calculation (but the transform itself is not really constant Q, instead the Q factor is actually variable/clamped), with musical tone scale, from E0 to D#10.
The filter accepts the following options:
Specify the video size for the output. It must be even. For the syntax of this option,
check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is 1920x1080
.
Set the output frame rate. Default value is 25
.
Set the bargraph height. It must be even. Default value is -1
which
computes the bargraph height automatically.
Set the axis height. It must be even. Default value is -1
which computes
the axis height automatically.
Set the sonogram height. It must be even. Default value is -1
which
computes the sonogram height automatically.
Set the fullhd resolution. This option is deprecated, use size, s
instead. Default value is 1
.
Specify the sonogram volume expression. It can contain variables:
the bar_v evaluated expression
the frequency where it is evaluated
the value of timeclamp option
and functions:
A-weighting of equal loudness
B-weighting of equal loudness
C-weighting of equal loudness.
Default value is 16
.
Specify the bargraph volume expression. It can contain variables:
the sono_v evaluated expression
the frequency where it is evaluated
the value of timeclamp option
and functions:
A-weighting of equal loudness
B-weighting of equal loudness
C-weighting of equal loudness.
Default value is sono_v
.
Specify the sonogram gamma. Lower gamma makes the spectrum more contrast,
higher gamma makes the spectrum having more range. Default value is 3
.
Acceptable range is [1, 7]
.
Specify the bargraph gamma. Default value is 1
. Acceptable range is
[1, 7]
.
Specify the bargraph transparency level. Lower value makes the bargraph sharper.
Default value is 1
. Acceptable range is [0, 1]
.
Specify the transform timeclamp. At low frequency, there is trade-off between
accuracy in time domain and frequency domain. If timeclamp is lower,
event in time domain is represented more accurately (such as fast bass drum),
otherwise event in frequency domain is represented more accurately
(such as bass guitar). Acceptable range is [0.002, 1]
. Default value is 0.17
.
Set attack time in seconds. The default is 0
(disabled). Otherwise, it
limits future samples by applying asymmetric windowing in time domain, useful
when low latency is required. Accepted range is [0, 1]
.
Specify the transform base frequency. Default value is 20.01523126408007475
,
which is frequency 50 cents below E0. Acceptable range is [10, 100000]
.
Specify the transform end frequency. Default value is 20495.59681441799654
,
which is frequency 50 cents above D#10. Acceptable range is [10, 100000]
.
This option is deprecated and ignored.
Specify the transform length in time domain. Use this option to control accuracy trade-off between time domain and frequency domain at every frequency sample. It can contain variables:
the frequency where it is evaluated
the value of timeclamp option.
Default value is 384*tc/(384+tc*f)
.
Specify the transform count for every video frame. Default value is 6
.
Acceptable range is [1, 30]
.
Specify the transform count for every single pixel. Default value is 0
,
which makes it computed automatically. Acceptable range is [0, 10]
.
Specify font file for use with freetype to draw the axis. If not specified, use embedded font. Note that drawing with font file or embedded font is not implemented with custom basefreq and endfreq, use axisfile option instead.
Specify fontconfig pattern. This has lower priority than fontfile. The
:
in the pattern may be replaced by |
to avoid unnecessary
escaping.
Specify font color expression. This is arithmetic expression that should return integer value 0xRRGGBB. It can contain variables:
the frequency where it is evaluated
the value of timeclamp option
and functions:
midi number of frequency f, some midi numbers: E0(16), C1(24), C2(36), A4(69)
red, green, and blue value of intensity x.
Default value is st(0, (midi(f)-59.5)/12);
st(1, if(between(ld(0),0,1), 0.5-0.5*cos(2*PI*ld(0)), 0));
r(1-ld(1)) + b(ld(1))
.
Specify image file to draw the axis. This option override fontfile and fontcolor option.
Enable/disable drawing text to the axis. If it is set to 0
, drawing to
the axis is disabled, ignoring fontfile and axisfile option.
Default value is 1
.
Set colorspace. The accepted values are:
Unspecified (default)
BT.709
FCC
BT.470BG or BT.601-6 625
SMPTE-170M or BT.601-6 525
SMPTE-240M
BT.2020 with non-constant luminance
Set spectrogram color scheme. This is list of floating point values with format
left_r|left_g|left_b|right_r|right_g|right_b
.
The default is 1|0.5|0|0|0.5|1
.
ffplay -f lavfi 'amovie=a.mp3, asplit [a][out1]; [a] showcqt [out0]'
ffplay -f lavfi 'amovie=a.mp3, asplit [a][out1]; [a] showcqt=fps=30:count=5 [out0]'
ffplay -f lavfi 'amovie=a.mp3, asplit [a][out1]; [a] showcqt=s=1280x720:count=4 [out0]'
sono_h=0
ffplay -f lavfi 'aevalsrc=0.1*sin(2*PI*55*t)+0.1*sin(4*PI*55*t)+0.1*sin(6*PI*55*t)+0.1*sin(8*PI*55*t), asplit[a][out1]; [a] showcqt [out0]'
ffplay -f lavfi 'aevalsrc=0.1*sin(2*PI*55*t)+0.1*sin(4*PI*55*t)+0.1*sin(6*PI*55*t)+0.1*sin(8*PI*55*t), asplit[a][out1]; [a] showcqt=timeclamp=0.5 [out0]'
bar_v=10:sono_v=bar_v*a_weighting(f)
bar_g=2:sono_g=2
tc=0.33:tlength='st(0,0.17); 384*tc / (384 / ld(0) + tc*f /(1-ld(0))) + 384*tc / (tc*f / ld(0) + 384 /(1-ld(0)))'
fontcolor='if(mod(floor(midi(f)+0.5),12), 0x0000FF, g(1))':fontfile=myfont.ttf
font='Courier New,Monospace,mono|bold'
axisfile=myaxis.png:basefreq=40:endfreq=10000
Convert input audio to video output representing frequency spectrum using Continuous Wavelet Transform and Morlet wavelet.
The filter accepts the following options:
Specify the video size for the output. For the syntax of this option,
check the (ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is 640x512
.
Set the output frame rate. Default value is 25
.
Set the frequency scale used. Allowed values are:
Default value is linear
.
Set the intensity scale used. Allowed values are:
Default value is log
.
Set the minimum frequency that will be used in output.
Default is 20
Hz.
Set the maximum frequency that will be used in output.
Default is 20000
Hz. The real frequency upper limit
depends on input audio’s sample rate and such will be enforced
on this value when it is set to value greater than Nyquist frequency.
Set the minimum intensity that will be used in output.
Set the maximum intensity that will be used in output.
Set the logarithmic basis for brightness strength when
mapping calculated magnitude values to pixel values.
Allowed range is from 0
to 1
.
Default value is 0.0001
.
Set the frequency deviation.
Lower values than 1
are more frequency oriented,
while higher values than 1
are more time oriented.
Allowed range is from 0
to 10
.
Default value is 1
.
Set the number of pixel output per each second in one row.
Allowed range is from 1
to 1024
.
Default value is 64
.
Set the output visual mode. Allowed values are:
Show magnitude.
Show only phase.
Show combination of magnitude and phase. Magnitude is mapped to brightness and phase to color.
Show unique color per channel magnitude.
Show unique color per stereo difference.
Default value is magnitude
.
Set the output slide method. Allowed values are:
Set the direction method for output slide method. Allowed values are:
Direction from left to right.
Direction from right to left.
Direction from up to down.
Direction from down to up.
Set the ratio of bargraph display to display size. Default is 0.
Set color rotation, must be in [-1.0, 1.0] range.
Default value is 0
.
Convert input audio to video output representing the audio power spectrum. Audio amplitude is on Y-axis while frequency is on X-axis.
The filter accepts the following options:
Specify size of video. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default is 1024x512
.
Set video rate. Default is 25
.
Set display mode. This set how each frequency bin will be represented.
It accepts the following values:
Default is bar
.
Set amplitude scale.
It accepts the following values:
Linear scale.
Square root scale.
Cubic root scale.
Logarithmic scale.
Default is log
.
Set frequency scale.
It accepts the following values:
Linear scale.
Logarithmic scale.
Reverse logarithmic scale.
Default is lin
.
Set window size. Allowed range is from 16 to 65536.
Default is 2048
Set windowing function.
It accepts the following values:
Default is hanning
.
Set window overlap. In range [0, 1]
. Default is 1
,
which means optimal overlap for selected window function will be picked.
Set time averaging. Setting this to 0 will display current maximal peaks.
Default is 1
, which means time averaging is disabled.
Specify list of colors separated by space or by ’|’ which will be used to draw channel frequencies. Unrecognized or missing colors will be replaced by white color.
Set channel display mode.
It accepts the following values:
Default is combined
.
Set minimum amplitude used in log
amplitude scaler.
Set data display mode.
It accepts the following values:
Default is magnitude
.
Set channels to use when processing audio. By default all are processed.
Convert stereo input audio to a video output, representing the spatial relationship between two channels.
The filter accepts the following options:
Specify the video size for the output. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is 512x512
.
Set window size. Allowed range is from 1024 to 65536. Default size is 4096.
Set window function.
It accepts the following values:
Default value is hann
.
Set output framerate.
Convert input audio to a video output, representing the audio frequency spectrum.
The filter accepts the following options:
Specify the video size for the output. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is 640x512
.
Specify how the spectrum should slide along the window.
It accepts the following values:
the samples start again on the left when they reach the right
the samples scroll from right to left
frames are only produced when the samples reach the right
the samples scroll from left to right
the samples start again on the right when they reach the left
Default value is replace
.
Specify display mode.
It accepts the following values:
all channels are displayed in the same row
all channels are displayed in separate rows
Default value is ‘combined’.
Specify display color mode.
It accepts the following values:
each channel is displayed in a separate color
each channel is displayed using the same color scheme
each channel is displayed using the rainbow color scheme
each channel is displayed using the moreland color scheme
each channel is displayed using the nebulae color scheme
each channel is displayed using the fire color scheme
each channel is displayed using the fiery color scheme
each channel is displayed using the fruit color scheme
each channel is displayed using the cool color scheme
each channel is displayed using the magma color scheme
each channel is displayed using the green color scheme
each channel is displayed using the viridis color scheme
each channel is displayed using the plasma color scheme
each channel is displayed using the cividis color scheme
each channel is displayed using the terrain color scheme
Default value is ‘channel’.
Specify scale used for calculating intensity color values.
It accepts the following values:
linear
square root, default
cubic root
logarithmic
4th root
5th root
Default value is ‘sqrt’.
Specify frequency scale.
It accepts the following values:
linear
logarithmic
Default value is ‘lin’.
Set saturation modifier for displayed colors. Negative values provide
alternative color scheme. 0
is no saturation at all.
Saturation must be in [-10.0, 10.0] range.
Default value is 1
.
Set window function.
It accepts the following values:
Default value is hann
.
Set orientation of time vs frequency axis. Can be vertical
or
horizontal
. Default is vertical
.
Set ratio of overlap window. Default value is 0
.
When value is 1
overlap is set to recommended size for specific
window function currently used.
Set scale gain for calculating intensity color values.
Default value is 1
.
Set which data to display. Can be magnitude
, default or phase
,
or unwrapped phase: uphase
.
Set color rotation, must be in [-1.0, 1.0] range.
Default value is 0
.
Set start frequency from which to display spectrogram. Default is 0
.
Set stop frequency to which to display spectrogram. Default is 0
.
Set upper frame rate limit. Default is auto
, unlimited.
Draw time and frequency axes and legends. Default is disabled.
Set dynamic range used to calculate intensity color values. Default is 120 dBFS. Allowed range is from 10 to 200.
Set upper limit of input audio samples volume in dBFS. Default is 0 dBFS. Allowed range is from -100 to 100.
Set opacity strength when using pixel format output with alpha component.
The usage is very similar to the showwaves filter; see the examples in that section.
showspectrum=s=1280x480:scale=log
ffplay
:
ffplay -f lavfi 'amovie=input.mp3, asplit [a][out1]; [a] showspectrum=mode=separate:color=intensity:slide=1:scale=cbrt [out0]'
Convert input audio to a single video frame, representing the audio frequency spectrum.
The filter accepts the following options:
Specify the video size for the output. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is 4096x2048
.
Specify display mode.
It accepts the following values:
all channels are displayed in the same row
all channels are displayed in separate rows
Default value is ‘combined’.
Specify display color mode.
It accepts the following values:
each channel is displayed in a separate color
each channel is displayed using the same color scheme
each channel is displayed using the rainbow color scheme
each channel is displayed using the moreland color scheme
each channel is displayed using the nebulae color scheme
each channel is displayed using the fire color scheme
each channel is displayed using the fiery color scheme
each channel is displayed using the fruit color scheme
each channel is displayed using the cool color scheme
each channel is displayed using the magma color scheme
each channel is displayed using the green color scheme
each channel is displayed using the viridis color scheme
each channel is displayed using the plasma color scheme
each channel is displayed using the cividis color scheme
each channel is displayed using the terrain color scheme
Default value is ‘intensity’.
Specify scale used for calculating intensity color values.
It accepts the following values:
linear
square root, default
cubic root
logarithmic
4th root
5th root
Default value is ‘log’.
Specify frequency scale.
It accepts the following values:
linear
logarithmic
Default value is ‘lin’.
Set saturation modifier for displayed colors. Negative values provide
alternative color scheme. 0
is no saturation at all.
Saturation must be in [-10.0, 10.0] range.
Default value is 1
.
Set window function.
It accepts the following values:
Default value is hann
.
Set orientation of time vs frequency axis. Can be vertical
or
horizontal
. Default is vertical
.
Set scale gain for calculating intensity color values.
Default value is 1
.
Draw time and frequency axes and legends. Default is enabled.
Set color rotation, must be in [-1.0, 1.0] range.
Default value is 0
.
Set start frequency from which to display spectrogram. Default is 0
.
Set stop frequency to which to display spectrogram. Default is 0
.
Set dynamic range used to calculate intensity color values. Default is 120 dBFS. Allowed range is from 10 to 200.
Set upper limit of input audio samples volume in dBFS. Default is 0 dBFS. Allowed range is from -100 to 100.
Set opacity strength when using pixel format output with alpha component.
ffmpeg
:
ffmpeg -i audio.flac -lavfi showspectrumpic=s=1024x1024 spectrogram.png
Convert input audio volume to a video output.
The filter accepts the following options:
Set video rate.
Set border width, allowed range is [0, 5]. Default is 1.
Set channel width, allowed range is [80, 8192]. Default is 400.
Set channel height, allowed range is [1, 900]. Default is 20.
Set fade, allowed range is [0, 1]. Default is 0.95.
Set volume color expression.
The expression can use the following variables:
Current max volume of channel in dB.
Current peak.
Current channel number, starting from 0.
If set, displays channel names. Default is enabled.
If set, displays volume values. Default is enabled.
Set orientation, can be horizontal: h
or vertical: v
,
default is h
.
Set step size, allowed range is [0, 5]. Default is 0, which means step is disabled.
Set background opacity, allowed range is [0, 1]. Default is 0.
Set metering mode, can be peak: p
or rms: r
,
default is p
.
Set display scale, can be linear: lin
or log: log
,
default is lin
.
In second.
If set to > 0., display a line for the max level
in the previous seconds.
default is disabled: 0.
The color of the max line. Use when dm
option is set to > 0.
default is: orange
Convert input audio to a video output, representing the samples waves.
The filter accepts the following options:
Specify the video size for the output. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is 600x240
.
Set display mode.
Available values are:
Draw a point for each sample.
Draw a vertical line for each sample.
Draw a point for each sample and a line between them.
Draw a centered vertical line for each sample.
Default value is point
.
Set the number of samples which are printed on the same column. A larger value will decrease the frame rate. Must be a positive integer. This option can be set only if the value for rate is not explicitly specified.
Set the (approximate) output frame rate. This is done by setting the option n. Default value is "25".
Set if channels should be drawn separately or overlap. Default value is 0.
Set colors separated by ’|’ which are going to be used for drawing of each channel.
Set amplitude scale.
Available values are:
Linear.
Logarithmic.
Square root.
Cubic root.
Default is linear.
Set the draw mode. This is mostly useful to set for high n.
Available values are:
Scale pixel values for each drawn sample.
Draw every sample directly.
Default value is scale
.
amovie=a.mp3,asplit[out0],showwaves[out1]
aevalsrc=sin(1*2*PI*t)*sin(880*2*PI*t):cos(2*PI*200*t),asplit[out0],showwaves=r=30[out1]
Convert input audio to a single video frame, representing the samples waves.
The filter accepts the following options:
Specify the video size for the output. For the syntax of this option, check the
(ffmpeg-utils)"Video size" section in the ffmpeg-utils manual.
Default value is 600x240
.
Set if channels should be drawn separately or overlap. Default value is 0.
Set colors separated by ’|’ which are going to be used for drawing of each channel.
Set amplitude scale.
Available values are:
Linear.
Logarithmic.
Square root.
Cubic root.
Default is linear.
Set the draw mode.
Available values are:
Scale pixel values for each drawn sample.
Draw every sample directly.
Default value is scale
.
Set the filter mode.
Available values are:
Use average samples values for each drawn sample.
Use peak samples values for each drawn sample.
Default value is average
.
ffmpeg
:
ffmpeg -i audio.flac -lavfi showwavespic=split_channels=1:s=1024x800 waveform.png
Delete frame side data, or select frames based on it.
This filter accepts the following options:
Set mode of operation of the filter.
Can be one of the following:
Select every frame with side data of type
.
Delete side data of type
. If type
is not set, delete all side
data in the frame.
Set side data type used with all modes. Must be set for select
mode. For
the list of frame side data types, refer to the AVFrameSideDataType
enum
in libavutil/frame.h. For example, to choose
AV_FRAME_DATA_PANSCAN
side data, you must specify PANSCAN
.
Synthesize audio from 2 input video spectrums, first input stream represents magnitude across time and second represents phase across time. The filter will transform from frequency domain as displayed in videos back to time domain as presented in audio output.
This filter is primarily created for reversing processed showspectrum
filter outputs, but can synthesize sound from other spectrograms too.
But in such case results are going to be poor if the phase data is not
available, because in such cases phase data need to be recreated, usually
it’s just recreated from random noise.
For best results use gray only output (channel
color mode in
showspectrum filter) and log
scale for magnitude video and
lin
scale for phase video. To produce phase, for 2nd video, use
data
option. Inputs videos should generally use fullframe
slide mode as that saves resources needed for decoding video.
The filter accepts the following options:
Specify sample rate of output audio, the sample rate of audio from which spectrum was generated may differ.
Set number of channels represented in input video spectrums.
Set scale which was used when generating magnitude input spectrum.
Can be lin
or log
. Default is log
.
Set slide which was used when generating inputs spectrums.
Can be replace
, scroll
, fullframe
or rscroll
.
Default is fullframe
.
Set window function used for resynthesis.
Set window overlap. In range [0, 1]
. Default is 1
,
which means optimal overlap for selected window function will be picked.
Set orientation of input videos. Can be vertical
or horizontal
.
Default is vertical
.
ffmpeg -i input.flac -lavfi showspectrum=mode=separate:scale=log:overlap=0.875:color=channel:slide=fullframe:data=magnitude -an -c:v rawvideo magnitude.nut ffmpeg -i input.flac -lavfi showspectrum=mode=separate:scale=lin:overlap=0.875:color=channel:slide=fullframe:data=phase -an -c:v rawvideo phase.nut ffmpeg -i magnitude.nut -i phase.nut -lavfi spectrumsynth=channels=2:sample_rate=44100:win_func=hann:overlap=0.875:slide=fullframe output.flac
Split input into several identical outputs.
asplit
works with audio input, split
with video.
The filter accepts a single parameter which specifies the number of outputs. If unspecified, it defaults to 2.
[in] split [out0][out1]
[in] asplit=3 [out0][out1][out2]
[in] split [splitout1][splitout2]; [splitout1] crop=100:100:0:0 [cropout]; [splitout2] pad=200:200:100:100 [padout];
ffmpeg
:
ffmpeg -i INPUT -filter_complex asplit=5 OUTPUT
Receive commands sent through a libzmq client, and forward them to filters in the filtergraph.
zmq
and azmq
work as a pass-through filters. zmq
must be inserted between two video filters, azmq
between two
audio filters. Both are capable to send messages to any filter type.
To enable these filters you need to install the libzmq library and
headers and configure FFmpeg with --enable-libzmq
.
For more information about libzmq see: http://www.zeromq.org/
The zmq
and azmq
filters work as a libzmq server, which
receives messages sent through a network interface defined by the
bind_address (or the abbreviation "b") option.
Default value of this option is tcp://localhost:5555. You may
want to alter this value to your needs, but do not forget to escape any
’:’ signs (see filtergraph escaping).
The received message must be in the form:
TARGET COMMAND [ARG]
TARGET specifies the target of the command, usually the name of the filter class or a specific filter instance name. The default filter instance name uses the pattern ‘Parsed_<filter_name>_<index>’, but you can override this by using the ‘filter_name@id’ syntax (see Filtergraph syntax).
COMMAND specifies the name of the command for the target filter.
ARG is optional and specifies the optional argument list for the given COMMAND.
Upon reception, the message is processed and the corresponding command is injected into the filtergraph. Depending on the result, the filter will send a reply to the client, adopting the format:
ERROR_CODE ERROR_REASON MESSAGE
MESSAGE is optional.
Look at tools/zmqsend for an example of a zmq client which can be used to send commands processed by these filters.
Consider the following filtergraph generated by ffplay
.
In this example the last overlay filter has an instance name. All other
filters will have default instance names.
ffplay -dumpgraph 1 -f lavfi " color=s=100x100:c=red [l]; color=s=100x100:c=blue [r]; nullsrc=s=200x100, zmq [bg]; [bg][l] overlay [bg+l]; [bg+l][r] overlay@my=x=100 "
To change the color of the left side of the video, the following command can be used:
echo Parsed_color_0 c yellow | tools/zmqsend
To change the right side:
echo Parsed_color_1 c pink | tools/zmqsend
To change the position of the right side:
echo overlay@my x 150 | tools/zmqsend
Below is a description of the currently available multimedia sources.
This is the same as movie source, except it selects an audio stream by default.
Generate an Audio/Video Sync Test.
Generated stream periodically shows flash video frame and emits beep in audio. Useful to inspect A/V sync issues.
It accepts the following options:
Set output video size. Default value is hd720
.
Set output video frame rate. Default value is 30
.
Set output audio sample rate. Default value is 44100
.
Set output audio beep amplitude. Default value is 0.7
.
Set output audio beep period in seconds. Default value is 3
.
Set output video flash delay in number of frames. Default value is 0
.
Enable cycling of video delays, by default is disabled.
Set stream output duration. By default duration is unlimited.
Set foreground/background/additional color.
Read audio and/or video stream(s) from a movie container.
It accepts the following parameters:
The name of the resource to read (not necessarily a file; it can also be a device or a stream accessed through some protocol).
Specifies the format assumed for the movie to read, and can be either the name of a container or an input device. If not specified, the format is guessed from movie_name or by probing.
Specifies the seek point in seconds. The frames will be output
starting from this seek point. The parameter is evaluated with
av_strtod
, so the numerical value may be suffixed by an IS
postfix. The default value is "0".
Specifies the streams to read. Several streams can be specified, separated by "+". The source will then have as many outputs, in the same order. The syntax is explained in the (ffmpeg)"Stream specifiers" section in the ffmpeg manual. Two special names, "dv" and "da" specify respectively the default (best suited) video and audio stream. Default is "dv", or "da" if the filter is called as "amovie".
Specifies the index of the video stream to read. If the value is -1, the most suitable video stream will be automatically selected. The default value is "-1". Deprecated. If the filter is called "amovie", it will select audio instead of video.
Specifies how many times to read the stream in sequence. If the value is 0, the stream will be looped infinitely. Default value is "1".
Note that when the movie is looped the source timestamps are not changed, so it will generate non monotonically increasing timestamps.
Specifies the time difference between frames above which the point is considered a timestamp discontinuity which is removed by adjusting the later timestamps.
Specifies the number of threads for decoding
Specify format options for the opened file. Format options can be specified as a list of key=value pairs separated by ’:’. The following example shows how to add protocol_whitelist and protocol_blacklist options:
ffplay -f lavfi "movie=filename='1.sdp':format_opts='protocol_whitelist=file,rtp,udp\:protocol_blacklist=http'"
It allows overlaying a second video on top of the main input of a filtergraph, as shown in this graph:
input -----------> deltapts0 --> overlay --> output ^ | movie --> scale--> deltapts1 -------+
movie=in.avi:seek_point=3.2, scale=180:-1, setpts=PTS-STARTPTS [over]; [in] setpts=PTS-STARTPTS [main]; [main][over] overlay=16:16 [out]
movie=/dev/video0:f=video4linux2, scale=180:-1, setpts=PTS-STARTPTS [over]; [in] setpts=PTS-STARTPTS [main]; [main][over] overlay=16:16 [out]
movie=dvd.vob:s=v:0+#0x81 [video] [audio]
Both movie and amovie support the following commands:
Perform seek using "av_seek_frame". The syntax is: seek stream_index|timestamp|flags
Get movie duration in AV_TIME_BASE units.
The FFmpeg developers.
For details about the authorship, see the Git history of the project
(https://git.ffmpeg.org/ffmpeg), e.g. by typing the command
git log
in the FFmpeg source directory, or browsing the
online repository at https://git.ffmpeg.org/ffmpeg.
Maintainers for the specific components are listed in the file MAINTAINERS in the source code tree.
This document was generated using makeinfo.