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, 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 in output the top half of the video is mirrored onto the bottom half.
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 the 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", a filter with no output pads is called a "sink".
A filtergraph can be represented using a textual representation, which is
recognized by the ‘-filter’/‘-vf’ and ‘-filter_complex’
options in ffmpeg and ‘-vf’ in ffplay, and by the
avfilter_graph_parse()/avfilter_graph_parse2() 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=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. 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 the following 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.
The name and arguments of the filter are optionally preceded and followed by a list of link labels. A link label allows 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 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.
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.
Follows a BNF description for the filtergraph syntax:
NAME ::= sequence of alphanumeric characters and '_' LINKLABEL ::= "[" NAME "]" LINKLABELS ::= LINKLABEL [LINKLABELS] FILTER_ARGUMENTS ::= sequence of chars (eventually quoted) FILTER ::= [LINKLABELS] NAME ["=" FILTER_ARGUMENTS] [LINKLABELS] FILTERCHAIN ::= FILTER [,FILTERCHAIN] FILTERGRAPH ::= [sws_flags=flags;] FILTERCHAIN [;FILTERGRAPH] |
Filtergraph description composition entails several levels of escaping. See (ffmpeg-utils)quoting_and_escaping 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" |
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
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 |
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.
Convert the input audio format to the specified formats.
This filter is deprecated. Use aformat instead.
The filter accepts a string of the form: "sample_format:channel_layout".
sample_format specifies the sample format, and can be a string or the corresponding numeric value defined in ‘libavutil/samplefmt.h’. Use ’p’ suffix for a planar sample format.
channel_layout specifies the channel layout, and can be a string or the corresponding number value defined in ‘libavutil/channel_layout.h’.
The special parameter "auto", signifies that the filter will automatically select the output format depending on the output filter.
aconvert=fltp:stereo |
aconvert=u8:auto |
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 ’|’. At least one delay greater than 0 should be provided. Unused delays will be silently ignored. If number of given delays is smaller than number of channels all remaining channels will not be delayed.
adelay=1500|0|500 |
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 loudnesses 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 |
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.
This filter accepts the following options:
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 |
eval=val(0)|-val(1) |
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 time for starting to apply the fade effect. Default is 0. The accepted syntax is:
[-]HH[:MM[:SS[.m...]]] [-]S+[.m...] |
See also the function av_parse_time().
If set this option is used instead of start_sample one.
Specify the duration for which the fade effect has to last. Default is 0. The accepted syntax is:
[-]HH[:MM[:SS[.m...]]] [-]S+[.m...] |
See also the function av_parse_time().
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.
If set this option is used instead of nb_samples one.
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
afade=t=in:ss=0:d=15 |
afade=t=out:st=875:d=25 |
Set output format constraints for the input audio. The framework will negotiate the most appropriate format to minimize conversions.
The filter accepts the following named 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)channel layout syntax for the required syntax.
If a parameter is omitted, all values are allowed.
For example to force the output to either unsigned 8-bit or signed 16-bit stereo:
aformat=sample_fmts=u8|s16:channel_layouts=stereo |
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
Specify the band-width of a filter in width_type units.
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.
For example
ffmpeg -i INPUT1 -i INPUT2 -i INPUT3 -filter_complex amix=inputs=3:duration=first:dropout_transition=3 OUTPUT |
will mix 3 input audio streams to a single output with the same duration as the first input and a dropout transition time of 3 seconds.
The filter accepts the following named parameters:
Number of inputs. If unspecified, it defaults to 2.
How to determine the end-of-stream.
Duration of longest input. (default)
Duration of shortest input.
Duration of first input.
Transition time, in seconds, for volume renormalization when an input stream ends. The default value is 2 seconds.
Pass the audio source unchanged to the output.
Pad the end of a audio stream with silence, this can be used together with -shortest to extend audio streams to the same length as the video stream.
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:
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 manual for the complete list of supported options.
aresample=44100 |
aresample=async=1000 |
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 signal 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.
A description of each shown parameter follows:
sequential number of the input frame, starting from 0
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.
presentation timestamp of the input frame in seconds
position of the frame in the input stream, -1 if this information in unavailable and/or meaningless (for example in case of synthetic audio)
sample format
channel layout
sample rate for the audio frame
number of samples (per channel) in the frame
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.
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.
The filter accepts the following option:
Short window length in seconds, used for peak and trough RMS measurement.
Default is 0.05 (50 miliseconds). Allowed range is [0.1 - 10].
A description of each shown parameter follows:
Mean amplitude displacement from zero.
Minimal sample level.
Maximal sample level.
Standard peak and RMS level measured in dBFS.
Peak and trough values for RMS level measured over a short window.
Standard ratio of peak to RMS level (note: not in dB).
Flatness (i.e. consecutive samples with the same value) of the signal at its peak levels (i.e. either Min level or Max level).
Number of occasions (not the number of samples) that the signal attained either Min level or Max level.
Forward two audio streams and control the order the buffers are forwarded.
The filter accepts the following options:
Set the expression deciding which stream should be forwarded next: if the result is negative, the first stream is forwarded; if the result is positive or zero, the second stream is forwarded. It can use the following variables:
number of buffers forwarded so far on each stream
number of samples forwarded so far on each stream
current timestamp of each stream
The default value is t1-t2, which means to always forward the stream
that has a smaller timestamp.
Stress-test amerge by randomly sending buffers on the wrong
input, while avoiding too much of a desynchronization:
amovie=file.ogg [a] ; amovie=file.mp3 [b] ; [a] [b] astreamsync=(2*random(1))-1+tanh(5*(t1-t2)) [a2] [b2] ; [a2] [b2] amerge |
Synchronize audio data with timestamps by squeezing/stretching it and/or dropping samples/adding silence when needed.
This filter is not built by default, please use aresample to do squeezing/stretching.
The filter accepts the following named parameters:
Enable stretching/squeezing the data to make it match the timestamps. Disabled by default. When disabled, time gaps are covered with silence.
Minimum difference between timestamps and audio data (in seconds) to trigger adding/dropping samples. Default value is 0.1. If you get non-perfect sync with this filter, try setting this parameter to 0.
Maximum compensation in samples per second. Relevant only with compensate=1. Default value 500.
Assume the first pts should be this value. The time base is 1 / sample rate. 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 silence if an audio stream starts after the video stream or to trim any samples with a negative pts due to encoder delay.
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, 2.0] range.
atempo=0.8 |
atempo=1.25 |
Trim the input so that the output contains one continuous subpart of the input.
This filter accepts the following options:
Specify time of the start of the kept section, 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.
Specify maximum duration of the output.
Number of the first sample that should be passed to output.
Number of the first sample that should be dropped.
‘start’, ‘end’, ‘duration’ are expressed as time duration specifications, check the "Time duration" section in the ffmpeg-utils 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 that 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 |
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
Specify the band-width of a filter in width_type units.
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
Specify the band-width of a filter in width_type units.
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
Determine how steep is the filter’s shelf transition.
Apply a biquad IIR filter with the given coefficients. Where b0, b1, b2 and a0, a1, a2 are the numerator and denominator coefficients respectively.
Remap input channels to new locations.
This filter accepts the following named parameters:
Channel layout of the output stream.
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.
If no mapping is present, the filter will implicitly map input channels to output channels preserving index.
For example, assuming a 5.1+downmix input MOV file
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.
To fix a 5.1 WAV improperly encoded in AAC’s native channel order
ffmpeg -i in.wav -filter 'channelmap=1|2|0|5|3|4:channel_layout=5.1' out.wav |
Split each channel in input audio stream into a separate output stream.
This filter accepts the following named parameters:
Channel layout of the input stream. Default is "stereo".
For example, assuming a stereo input MP3 file
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.
To split a 5.1 WAV file into per-channel files
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 |
Compress or expand audio dynamic range.
A description of the accepted options follows.
Set 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.
Typical value for attack is 0.3 seconds and for decay 0.8
seconds.
Set list of points for transfer function, specified in dB relative to maximum
possible signal amplitude.
Each key points list need to be defined using the following syntax:
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.
Set amount for which the points at where adjacent line segments on the
transfer function meet will be rounded. Defaults is 0.01.
Set additional gain in dB to be applied at all points on the transfer function
and allows easy adjustment of the overall gain.
Default is 0.
Set 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. Default is 0.
Set delay in seconds. Default is 0. The input audio
is analysed 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.
compand=.3 .3:1 1:-90/-60 -60/-40 -40/-30 -20/-20:6:0:-90:0.2 |
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 |
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
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.
equalizer=f=1000:width_type=h:width=200:g=-10 |
equalizer=f=1000:width_type=q:width=1:g=2,equalizer=f=100:width_type=q:width=2:g=-5 |
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
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.
Join multiple input streams into one multi-channel stream.
The filter accepts the following named parameters:
Number of input streams. Defaults to 2.
Desired output channel layout. 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 those 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.
E.g. to join 3 inputs (with properly set channel layouts)
ffmpeg -i INPUT1 -i INPUT2 -i INPUT3 -filter_complex join=inputs=3 OUTPUT |
To 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.
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 the function
av_parse_time() for the accepted format, also check the "Time duration"
section in the ffmpeg-utils manual.
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.
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 |
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.
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
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.
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 remap efficiently 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=1: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.
Convert the audio sample format, sample rate and channel layout. This filter is not meant to be used directly.
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 filter accepts the following options:
Set silence duration until notification (default is 2 seconds).
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.
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 - |
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
Determine how steep is the filter’s shelf transition.
Adjust the input audio volume.
The filter accepts the following options:
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 |
Default value for volume is "1.0".
Set the mathematical precision.
This determines which input sample formats will be allowed, which affects the precision of the volume scaling.
8-bit fixed-point; limits input sample format to U8, S16, and S32.
32-bit floating-point; limits input sample format to FLT. (default)
64-bit floating-point; limits input sample format to DBL.
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
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. The input is not modified. 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/asrc_abuffer.h’.
It accepts the following named parameters:
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 corresponging 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 the function
av_parse_time() for the accepted format.
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)" |
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:
Specify 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.
Specify the sample rate, and defaults to 44100.
Set the number of samples per requested frames.
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 the flite library is 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.speech.cs.cmu.edu/flite/
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, default is 1024.
sine |
sine=220:4:d=5 sine=f=220:b=4:d=5 sine=frequency=220:beep_factor=4:duration=5 |
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 to be employed 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.
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] |
Since this filter is designed for reconstruction, it operates on frame sequences without considering timestamps, and terminates when either input reaches end of stream. This will cause problems if your encoding pipeline drops frames. If you’re trying to apply an image as an overlay to a video stream, consider the overlay filter instead.
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.
Compute the bounding box for the non-black pixels in the input frame luminance plane.
This filter computes the bounding box containing all the pixels with a luminance 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 luminance value. Default is 16.
Detect video intervals that are (almost) completely black. Can be useful to detect chapter transitions, commercials, or invalid recordings. Output lines contains the time for the start, end and duration of the detected black interval expressed in seconds.
In order to display the output lines, you need to set the loglevel at least to the AV_LOG_INFO value.
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 luminance value for which a pixel is considered "black". The provided value is scaled according to the following equation:
absolute_threshold = luminance_minimum_value + pixel_black_th * luminance_range_size |
luminance_range_size and luminance_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.
The filter accepts the following options:
Set the percentage of the pixels that have to be below the threshold, defaults
to 98.
Set the threshold below which a pixel value is considered black, defaults to
32.
Blend two video frames into each other.
It takes two input streams and outputs one stream, the first input is the "top" layer and second input is "bottom" layer. Output terminates when shortest 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:
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 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.
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).
Force termination when the shortest input terminates. Default is 0.
Continue applying the last bottom frame after the end of the stream. A value of
0 disable the filter after the last frame of the bottom layer is reached.
Default is 1.
blend=all_expr='A*(if(gte(T,10),1,T/10))+B*(1-(if(gte(T,10),1,T/10)))' |
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)' |
Apply boxblur algorithm to the input video.
The filter accepts the following options:
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
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 |
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.
colorbalance=rs=.3 |
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].
colorchannelmixer=.3:.4:.3:0:.3:.4:.3:0:.3:.4:.3 |
colorchannelmixer=.393:.769:.189:0:.349:.686:.168:0:.272:.534:.131 |
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
BT.601
SMPTE-240M
FCC
For example to convert from BT.601 to SMPTE-240M, use the command:
colormatrix=bt601:smpte240m |
Copy the input source unchanged to the output. Mainly useful for testing purposes.
Crop the input video to given dimensions.
The filter accepts the following options:
Width of the output video. It defaults to iw.
This expression is evaluated only once during the filter
configuration.
Height of the output video. It defaults to ih.
This expression is evaluated only once during the filter
configuration.
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.
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.
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
same as in_w and in_h
the output (cropped) width and height
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 input frame, starting from 0
the position in the file of the input frame, NAN if unknown
timestamp expressed in seconds, 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) |
Auto-detect crop size.
Calculate necessary cropping parameters and prints the recommended parameters through the logging system. The detected dimensions correspond to the non-black area of the input video.
The filter accepts the following options:
Set higher black value threshold, which can be optionally specified from nothing (0) to everything (255). An intensity value greater to the set value is considered non-black. Default value is 24.
Set the value for which the width/height should be divisible by. 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. Default value is 16.
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 return the largest area encountered during playback.
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 (using a natural cubic spline interpolation) 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. If the computed curves happened to go outside the vector spaces, the values will be clipped accordingly.
If there is no key point defined in x=0, the filter will automatically
insert a (0;0) point. In the same way, if there is no key point defined
in x=1, the filter will automatically insert a (1;1) point.
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 (.asv) to import the settings from.
To avoid some filtergraph syntax conflicts, each key points list need to be
defined using the following syntax: x0/y0 x1/y1 x2/y2 ....
curves=blue='0.5/0.58' |
curves=r='0/0.11 .42/.51 1/0.95':g='0.50/0.48':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.asv':green='0.45/0.53' |
Denoise frames using 2D DCT (frequency domain filtering).
This filter is not designed for real time and can be extremely slow.
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. Each block is of size
16x16. 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 allow to process the whole input width or height, a warning will be displayed and according borders won’t be denoised.
Default value is 15.
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.
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)' |
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.
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).
This filter accepts the following options:
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.
Specify the thickness of the fuzzy edge of the rectangle (added to w and h). The default value is 4.
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.
delogo=x=0:y=0:w=100:h=77:band=10 |
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.
If set to 1, specify using OpenCL capabilities, only available if
FFmpeg was configured with --enable-opencl. Default value is 0.
Draw a colored box on the input image.
This filter accepts the following options:
The expressions which specify the top left corner coordinates of the box. Default to 0.
The expressions which specify the width and height of the box, if 0 they are interpreted as the input width and height. Default to 0.
Specify the color of the box to write. For the general syntax of this option,
check the "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. Default value is 3.
See below for the list of accepted constants.
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.
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=max |
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 grid on the input image.
This filter accepts the following options:
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 "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.
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 text string or text from specified file on top of video using the libfreetype library.
To enable compilation of this filter you need to configure FFmpeg with
--enable-libfreetype.
The description of the accepted parameters follows.
Used to draw a box around text using background color. Value should be either 1 (enable) or 0 (disable). The default value of box is 0.
The color to be used for drawing box around text. For the syntax of this option, check the "Color" section in the ffmpeg-utils manual.
The default value of boxcolor is "white".
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 "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.
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 "Color" section in the ffmpeg-utils manual.
The default value of fontcolor is "black".
The font file to be used for drawing text. Path must be included. This parameter is mandatory.
The font size to be used for drawing text. The default value of fontsize is 16.
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 "Color" section in the ffmpeg-utils manual.
The default value of shadowcolor is "black".
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. 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).
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.
If set to 1, the textfile will be reloaded before each frame. Be sure to update it atomically, or it may be read partially, or even fail.
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 number of input frame, starting from 0
return a random number included between min and max
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
each other, so you can for example specify y=x/dar.
If libavfilter was built with --enable-fontconfig, then
‘fontfile’ can be a fontconfig pattern or omitted.
If ‘expansion’ is set to strftime,
the filter recognizes strftime() sequences in the provided text and
expands them accordingly. Check the documentation of strftime(). This
feature is deprecated.
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.
Sequence 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 avoids these problems.
The following functions are available:
expr, eThe 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.
gmtimeThe time at which the filter is running, expressed in UTC. It can accept an argument: a strftime() format string.
localtimeThe time at which the filter is running, expressed in the local time zone. It can accept an argument: a strftime() format string.
metadataFrame metadata. It must take one argument specifying metadata key.
n, frame_numThe frame number, starting from 0.
pict_typeA 1 character description of the current picture type.
ptsThe timestamp of the current frame, in seconds, with microsecond accuracy.
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-line_h)/2" |
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='fontfile=FreeSans.ttf:text=%{localtime:%a %b %d %Y}'
|
For more information about libfreetype, check: http://www.freetype.org/.
For more information about fontconfig, check: http://freedesktop.org/software/fontconfig/fontconfig-user.html.
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 choosen 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.
Example:
edgedetect=low=0.1:high=0.4 |
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 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.
Apply fade-in/out effect to input video.
This filter accepts the following options:
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 frame for starting to apply the fade effect. Default is 0.
The number of frames 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’. 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.
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 |
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).
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 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.
This filter accepts the following options:
Output field order. Valid values are tff for top field first or bff for bottom field first.
Default value is ‘tff’.
Transformation is achieved 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.
This filter 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.
This filter is mainly useful when auto-inserted by the libavfilter framework.
The filter does not take parameters.
Convert the input video to one of the specified pixel formats. Libavfilter will try to pick one that is supported for the input to the next filter.
This filter accepts the following parameters:
A ’|’-separated list of pixel format names, for example "pix_fmts=yuv420p|monow|rgb24".
format=pix_fmts=yuv420p |
Convert the input video to any of the formats in the list
format=pix_fmts=yuv420p|yuv444p|yuv410p |
Convert the video to specified constant frame rate by duplicating or dropping frames as necessary.
This filter accepts the following named parameters:
Desired output frame rate. The default is 25.
Rounding method.
Possible values are:
zero round towards 0
round away from 0
round towards -infinity
round towards +infinity
round to nearest
The default is near.
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.
Alternatively, the options can be specified as a flat string: fps[: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.
This filter accepts the following named parameters:
Desired packing format. Supported values are:
Views are next to each other (default).
Views are on top of each other.
Views are packed by line.
Views are eacked by column.
Views are temporally interleaved.
Some examples follow:
# 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 |
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.
Apply a frei0r effect to the input video.
To enable compilation of this filter you need to install the frei0r
header and configure FFmpeg with --enable-frei0r.
This filter accepts the following options:
The name to the frei0r effect to load. If the environment variable
FREI0R_PATH is defined, the frei0r effect is searched in each one of the
directories specified by the colon separated list in FREIOR_PATH,
otherwise in the standard frei0r paths, which are 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 (whose values are specified with "y" and "n"), a double, a color (specified by the syntax R/G/B, (R, G, and B being float numbers from 0.0 to 1.0) or by a color description specified in the "Color" section in the ffmpeg-utils manual), a position (specified by the syntax X/Y, X and Y being float numbers) and a string.
The number and kind 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
The filter accepts the following options:
Set the luminance 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 luminance 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 luminance 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.
For functions, if x and y are outside the area, the value will be automatically clipped to the closer edge.
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)' |
Fix the banding artifacts that are sometimes introduced into nearly flat regions by truncation to 8bit color depth. Interpolate the gradients that should go where the bands are, and dither them.
This filter 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.
This filter accepts the following options:
The maximum amount by which the filter will change any one pixel. Also the threshold for detecting nearly flat regions. Acceptable values range from .51 to 64, 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, 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]
3.5 strength and radius of 8:
gradfun=3.5:8 |
gradfun=radius=8 |
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:
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).
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.
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 distribution of color components in an image.
The filter accepts the following options:
Set histogram mode.
It accepts the following values:
standard histogram that display 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 current frame. Bellow each graph is color component scale meter.
chroma values in vectorscope, if brighter more such chroma values are distributed in an image. Displays chroma values (U/V color placement) in two dimensional graph (which is called a vectorscope). It can be used to read of the hue and saturation of the current frame. At a same time it is a histogram. The whiter a pixel in the vectorscope, the more pixels of the input frame correspond to that pixel (that is the more pixels have this chroma value). The V component is displayed on the horizontal (X) axis, with the leftmost side being V = 0 and the rightmost side being V = 255. The U component is displayed on the vertical (Y) axis, with the top representing U = 0 and the bottom representing U = 255.
The position of a white pixel in the graph corresponds to the chroma value of a pixel of the input clip. So the graph can be used to read of the hue (color flavor) and the saturation (the dominance of the hue in the color). As the hue of a color changes, it moves around the square. At the center of the square, the saturation is zero, which means that the corresponding pixel has no color. If you increase the amount of a specific color, while leaving the other colors unchanged, the saturation increases, and you move towards the edge of the square.
chroma values in vectorscope, similar as color but actual chroma values
are displayed.
per row/column color component graph. In row mode graph in the left side represents color component value 0 and right side represents value = 255. In column mode top side represents color component value = 0 and bottom side represents value = 255.
Default value is levels.
Set height of level in levels. Default value is 200.
Allowed range is [50, 2048].
Set height of color scale in levels. Default value is 12.
Allowed range is [0, 40].
Set step for waveform mode. Smaller values are useful to find out how much
of same luminance values across input rows/columns are distributed.
Default value is 10. Allowed range is [1, 255].
Set mode for waveform. Can be either row, or column.
Default is row.
Set mirroring mode for waveform. 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
0 (unmirrored).
Set display mode for waveform and levels.
It accepts the following values:
Display separate graph for the color components side by side in
row waveform mode or one below other in column waveform mode
for waveform histogram mode. For levels histogram mode
per color component graphs are placed one bellow other.
This display mode in waveform histogram 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 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 make by making adjustments to level the
three waveforms.
Presents information that’s identical to that in the parade, except
that the graphs representing color components are superimposed directly
over one another.
This display mode in waveform histogram mode can make it easier to spot
the relative differences or similarities in overlapping areas of the color
components that are supposed to be identical, such as neutral whites, grays,
or blacks.
Default is parade.
Set mode for levels. Can be either linear, or logarithmic.
Default is linear.
ffplay -i input -vf histogram |
High precision/quality 3d denoise filter. This filter 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 float number which specifies spatial luma strength, defaults to 4.0
a non-negative float number which specifies spatial chroma strength, defaults to 3.0*luma_spatial/4.0
a float number which specifies luma temporal strength, defaults to 6.0*luma_spatial/4.0
a float number which specifies chroma temporal strength, defaults to luma_tmp*chroma_spatial/luma_spatial
Modify the hue and/or the saturation of the input.
This filter accepts the following options:
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.
Detect video interlacing type.
This filter tries to detect if the input is interlaced or progressive, top or bottom field first.
The filter accepts the following options:
Set interlacing threshold.
Set progressive threshold.
Deinterleave or interleave fields.
This filter allows 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.
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:
determines whether the interlaced frame is taken from the even (tff - default) or odd (bff) lines of the progressive frame.
Enable (default) or disable the vertical lowpass filter to avoid twitter interlacing and reduce moire patterns.
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 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
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.
Compute a look-up table for binding each pixel component input value to an output value, and apply it to input video.
lutyuv applies a lookup table to a YUV input video, lutrgb to an RGB input video.
These filters accept the following options:
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/luminance 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
input value for the pixel component
the input value clipped in the minval-maxval range
maximum value for the pixel component
minimum value for the pixel component
the negated value for the pixel component value clipped in the minval-maxval range , it corresponds to the expression "maxval-clipval+minval"
the computed value in val clipped in the minval-maxval range
the computed gamma correction value of the pixel component value clipped in the minval-maxval range, corresponds to the expression "pow((clipval-minval)/(maxval-minval)\,gamma)*(maxval-minval)+minval"
All expressions default to "val".
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)' |
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.
[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 |
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.
Apply an MPlayer filter to the input video.
This filter provides a wrapper around some of the filters of MPlayer/MEncoder.
This wrapper is considered experimental. Some of the wrapped filters may not work properly and we may drop support for them, as they will be implemented natively into FFmpeg. Thus you should avoid depending on them when writing portable scripts.
The filter accepts the parameters: filter_name[:=]filter_params
filter_name is the name of a supported MPlayer filter, filter_params is a string containing the parameters accepted by the named filter.
The list of the currently supported filters follows:
The parameter syntax and behavior for the listed filters are the same of the corresponding MPlayer filters. For detailed instructions check the "VIDEO FILTERS" section in the MPlayer manual.
mp=eq2=1.0:2:0.5 |
See also mplayer(1), http://www.mplayerhq.hu/.
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 unregarding the number of previous sequentially dropped 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.
Negate input video.
This filter accepts an integer in input, if non-zero it negates the alpha component (if available). The default value in input is 0.
Force libavfilter not to use any of the specified pixel formats for the input to the next filter.
This filter accepts the following parameters:
A ’|’-separated list of pixel format names, for example "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 |
Pass the video source unchanged to the output.
Apply video transform using libopencv.
To enable this filter install libopencv library and headers and
configure FFmpeg with --enable-libopencv.
This filter 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://opencv.willowgarage.com/documentation/c/image_filtering.html
Follows the list of supported libopencv filters.
Dilate an image by using a specific structuring element.
This filter 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, and can be one of the values "rect", "cross", "ellipse", "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.
Follow some example:
# use the default values ocv=dilate # dilate using a structuring element with a 5x5 cross, iterate two times ocv=filter_name=dilate:filter_params=5x5+2x2/cross|2 # read the shape from the file diamond.shape, iterate two times # the file diamond.shape may contain a pattern of characters like this: # * # *** # ***** # *** # * # the specified cols and rows are ignored (but not the anchor point coordinates) ocv=dilate:0x0+2x2/custom=diamond.shape|2 |
Erode an image by using a specific structuring element.
This filter corresponds to the libopencv function cvErode.
The filter 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 can be one of the following values: "blur", "blur_no_scale", "median", "gaussian", "bilateral". The default value is "gaussian".
param1, param2, param3, and param4 are parameters whose meanings depend on smooth type. param1 and param2 accept integer positive values or 0, param3 and param4 accept float 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.
Overlay one video on top of another.
It takes two inputs and one output, the first input is the "main" video on which the second input is overlayed.
This filter accepts the following parameters:
A description of the accepted options follows.
Set the expression for the x and y coordinates of the overlayed 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).
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’.
If set to 1, force the output to terminate when the shortest input terminates. Default value is 0.
Set the format for the output video.
It accepts the following values:
force YUV420 output
force YUV444 output
force RGB output
Default value is ‘yuv420’.
If set to 1, force the filter to accept inputs in the RGB color space. Default value is 0. This option is deprecated, use ‘format’ instead.
If set to 1, force the filter to draw the last overlay frame over the main input until the end of the stream. A value of 0 disables this behavior. Default value is 1.
The ‘x’, and ‘y’ expressions can contain the following parameters.
main input width and height
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
timestamp expressed in seconds, NAN if the input timestamp is unknown
Note that the n, pos, 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 it does the example for the movie filter.
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] " |
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] |
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 given coordinates x, y.
This filter 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 an expression for the offsets where to place the input image in 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.
Specify the color of the padded area. For the syntax of this option, check the "Color" section in the ffmpeg-utils manual.
The default value of color is "black".
The value for the width, height, x, and y options are expressions containing the following constants:
the input video width and height
same as in_w and in_h
the output width and height, that is the size of the padded area as specified by the width and height expressions
same as out_w and out_h
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
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" |
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.
The expressions can use the following variables:
the width and height of video frame.
Set interpolation for perspective correction.
It accepts the following values:
Default value is ‘linear’.
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.
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.
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 luminance filtering only (no chrominance).
Do chrominance filtering only (no luminance).
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 luminance 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 |
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.
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
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.
For example:
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’.
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 ... |
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.
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 generalsyntax of this option, check the "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 heigth
the output width and heigth, 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 0.1-4.0. 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 0.1-2.0 range.
Set chroma maximum difference between pixels to still be considered, must be a value in the 0.1-100.0 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)scaler_options for the complete list of scaler options.
Set the output video dimension expression. Default value is the input dimension.
If the value is 0, the input width is used for the output.
If one of the values is -1, the scale filter will use a value that maintains the aspect ratio of the input image, calculated from the other specified dimension. If both of them are -1, the input size is used
If one of the values is -n with n > 1, the scale filter will also 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.
See below for the list of accepted constants for use in the dimension expression.
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)sws_flags for the complete list of values. If not explictly specified the filter applies the default flags.
Set the video size. For the syntax of this option, check the "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 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.
The values of the ‘w’ and ‘h’ options are expressions containing the following constants:
the input width and height
same as in_w and in_h
the output (scaled) width and height
same as out_w and out_h
same as iw / ih
input sample aspect ratio
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.
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' |
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.
The filters accept the following options:
setdar only), sar (setsar only)’Set the aspect ratio used by the filter.
The parameter can be a floating point number string, an expression, or
a string of the form num:den, where num and
den are the numerator and denominator of the aspect ratio. If
the parameter is not specified, it is assumed the value "0".
In case the form "num:den" is used, the : character
should be escaped.
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 corresponding mathematical approximated values for e (euler number), pi (greek PI), phi (golden ratio)
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
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 setdar=dar=1.77777 |
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.
Show a line containing various information for each input video frame. The input video is not modified.
The shown line contains a sequence of key/value pairs of the form key:value.
A description of each shown parameter follows:
sequential number of the input frame, starting from 0
Presentation TimeStamp of the input frame, expressed as a number of time base units. The time base unit depends on the filter input pad.
Presentation TimeStamp of the input frame, expressed as a number of seconds
position of the frame in the input stream, -1 if this information in unavailable and/or meaningless (for example in case of synthetic video)
pixel format name
sample aspect ratio of the input frame, expressed in the form num/den
size of the input frame. For the syntax of this option, check the "Video size" section in the ffmpeg-utils manual.
interlaced mode ("P" for "progressive", "T" for top field first, "B" for bottom field first)
1 if the frame is a key frame, 0 otherwise
picture type of the input frame ("I" for an I-frame, "P" for a
P-frame, "B" for a B-frame, "?" for unknown type).
Check also the documentation of the AVPictureType enum and of
the av_get_picture_type_char function defined in
‘libavutil/avutil.h’.
Adler-32 checksum (printed in hexadecimal) of all the planes of the input frame
Adler-32 checksum (printed in hexadecimal) of each plane of the input frame, expressed in the form "[c0 c1 c2 c3]"
Blur the input video without impacting the outlines.
The filter 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 1.0.
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 1.0.
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 0.
If a chroma option is not explicitly set, the corresponding luma value is set.
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)
Default value is ‘sbsl’.
Set stereoscopic image format of output.
Available values for output image formats are all the input formats as well as:
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)
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)
mono output (left eye only)
mono output (right eye only)
Default value is ‘arcd’.
stereo3d=sbsl:aybd |
stereo3d=abl:sbsr |
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).
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 "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 subtitles input character encoding. subtitles filter only. Only
useful if not UTF-8.
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 |
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 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
|
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.
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.
thumbnail=50 |
ffmpeg:
ffmpeg -i in.avi -vf thumbnail,scale=300:200 -frames:v 1 out.png |
Tile several successive frames together.
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 "Video size" section in the ffmpeg-utils manual.
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.
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.
Specify the color of the unused areaFor the syntax of this option, check the "Color" section in the ffmpeg-utils manual. The default value of color is "black".
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 accomodate 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.
Only output even frames, odd frames are dropped, generating a frame with unchanged height at half frame rate.
Only output odd frames, even frames are dropped, generating a frame with unchanged height at half frame rate.
Expand each frame to full height, but pad alternate lines with black, generating a frame with double height at the same input frame rate.
Interleave the upper field from odd frames with the lower field from even frames, generating a frame with unchanged height at half frame rate.
Interleave the lower field from odd frames with the upper field from even frames, generating a frame with unchanged height at half frame rate.
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.
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 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.
Vertical low-pass filtering can only be enabled for ‘mode’ interleave_top and interleave_bottom.
Transpose rows with columns in the input video and optionally flip it.
This filter accepts the following options:
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 |
Trim the input so that the output contains one continuous subpart of the input.
This filter accepts the following options:
Specify 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 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.
Same as start, except this option sets the start timestamp in timebase units instead of seconds.
Same as end, except this option sets the end timestamp in timebase units instead of seconds.
Specify maximum duration of the output.
Number of the first frame that should be passed to output.
Number of the first frame that should be dropped.
‘start’, ‘end’, ‘duration’ are expressed as time duration specifications, check the "Time duration" section in the ffmpeg-utils manual.
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 that the output timestamps 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 |
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 63, default value is 5.
Set the luma matrix vertical size. It must be an odd integer between 3 and 63, default value is 5.
Set the luma effect strength. It can be a float 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 63, default value is 5.
Set the chroma matrix vertical size. It must be an odd integer between 3 and 63, default value is 5.
Set the chroma effect strength. It can be a float 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.
If set to 1, specify using OpenCL capabilities, only available if
FFmpeg was configured with --enable-opencl. Default value is 0.
All parameters are optional and default to the equivalent of the string ’5:5:1.0:5:5:0.0’.
unsharp=luma_msize_x=7:luma_msize_y=7:luma_amount=2.5 |
unsharp=7:7:-2:7:7:-2 |
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.
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 values 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 previsou 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 |
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 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 |
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.
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’.
Specify which frames to deinterlace. Accept one of the following values:
Deinterlace all frames,
Only deinterlace frames marked as interlaced.
Default value is ‘all’.
Deinterlace the input video ("yadif" means "yet another deinterlacing filter").
This filter accepts the following options:
The interlacing mode to adopt, accepts one of the following values:
output 1 frame for each frame
output 1 frame for each field
like send_frame but skip spatial interlacing check
like send_field but skip spatial interlacing check
Default value is send_frame.
The picture field parity assumed for the input interlaced video, accepts one of the following values:
assume top field first
assume bottom field first
enable automatic detection
Default value is auto.
If interlacing is unknown or decoder does not export this information,
top field first will be assumed.
Specify which frames to deinterlace. Accept one of the following values:
deinterlace all frames
only deinterlace frames marked as interlaced
Default value is all.
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/vsrc_buffer.h’.
This source accepts the following options:
Specify the size (width and height) of the buffered video frames. For the syntax of this option, check the "Video size" section in the ffmpeg-utils manual.
Input video width.
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.
Specify the sample aspect ratio assumed by the video frames.
Specify the optional parameters to be used for the scale filter which is automatically inserted when an input change is detected in the input size or format.
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[:sws_param]
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 "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 |
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 cound 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 "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 float number or a valid video frame rate abbreviation. The default value is "25".
Set the video duration of the sourced video. The accepted syntax is:
[-]HH:MM:SS[.m...] [-]S+[.m...] |
See also the function av_parse_time().
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:
Default value is "all", which will cycle through the list of all tests.
For example the following:
testsrc=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 options:
The size of the video to generate. For the syntax of this option, check the "Video size" section in the ffmpeg-utils manual.
Framerate of the generated video, 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 section frei0r in the description of the video filters.
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 overlayed 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 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 "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 "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 color source provides an uniformly colored 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 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 sources accept the following options:
Specify the color of the source, only available in the color
source. For the syntax of this option, check the "Color" section in the
ffmpeg-utils manual.
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 size of the sourced video. For the syntax of this option, check the "Video size" section in the ffmpeg-utils manual. The default value is "320x240".
This option is not available with the haldclutsrc filter.
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 float number or a valid video frame rate abbreviation. The default value is "25".
Set the sample aspect ratio of the sourced video.
Set the video duration of the sourced video. The accepted syntax is:
[-]HH[:MM[:SS[.m...]]] [-]S+[.m...] |
See also the function av_parse_time().
If not specified, or the expressed duration is negative, the video is supposed to be generated forever.
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.
For example the following:
testsrc=duration=5.3:size=qcif:rate=10 |
will generate a video with a duration of 5.3 seconds, with size 176x144 and a frame rate of 10 frames per second.
The following graph description will generate a red source with an opacity of 0.2, with size "qcif" and a frame rate of 10 frames per second.
color=c=red@0.2:s=qcif:r=10 |
If the input content is to be ignored, nullsrc can be used. The
following command generates noise in the luminance 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.
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 a 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 to be employed in analysis / debugging tools.
Below is a description of the currently available multimedia filters.
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 monoaural 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.
Default value is ‘lissajous’.
Set the video size for the output. For the syntax of this option, check the "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 and blue contrast. Default values are 40, 160 and 80.
Allowed range is [0, 255].
Specify the red, green and blue fade. Default values are 15, 10 and 5.
Allowed range is [0, 255].
Set the zoom factor. Default value is 1. Allowed range is [1, 10].
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]'
|
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 video 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 as input and outputs
it unchanged. 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 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).
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 "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:
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.
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 a 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 drop 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 will send 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.
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 |
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.
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, NAN if undefined
timebase of the input timestamps
the PTS (Presentation TimeStamp) of the filtered video frame, expressed in TB units, NAN if undefined
the PTS (Presentation TimeStamp) of the filtered video frame, expressed in seconds, NAN if undefined
the PTS of the previously filtered video frame, NAN if undefined
the PTS of the last previously filtered video frame, NAN if undefined
the PTS of the last previously selected video frame, NAN if undefined
the PTS of the first video frame in the video, NAN if undefined
the time of the first video frame in the video, NAN if undefined
the type of the filtered frame, can assume one of the following values:
the frame interlace type, 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
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)
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 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 |
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.
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 |
# 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)
audio sample rate
the PTS of the first frame
the time in seconds of the first frame
tell if 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
previous input PTS
previous input time in seconds
previous output PTS
previous output time in seconds
wallclock (RTC) time in microseconds. This is deprecated, use time(0) instead.
wallclock (RTC) time at the start of the movie in microseconds
timebase of the input timestamps
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 |
Set the timebase to use for the output frames timestamps. It is mainly useful for testing timebase configuration.
This filter accepts the following options:
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 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 "Video size" section in the ffmpeg-utils manual. Default value is
640x512.
Specify if the spectrum should slide along the window. Default value is
0.
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 is displayed using the same 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
Default value is ‘sqrt’.
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:
No samples pre-processing (do not expect this to be faster)
Hann window
Hamming window
Blackman window
Default value is hann.
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 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 "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.
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".
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] |
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.
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’ option.
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.
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
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=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 |
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.
Read audio and/or video stream(s) from a movie container.
This filter accepts the following options:
The name of the resource to read (not necessarily a file but also 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. 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 “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 best suited video stream will be automatically selected. 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 less than 1, the stream will be read again and again. 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.
This filter allows to overlay a second video on top of 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] |
ffmpeg, ffplay, ffprobe, libavfilter
The FFmpeg developers.
For details about the authorship, see the Git history of the project
(git://source.ffmpeg.org/ffmpeg), e.g. by typing the command
git log in the FFmpeg source directory, or browsing the
online repository at http://source.ffmpeg.org.
Maintainers for the specific components are listed in the file ‘MAINTAINERS’ in the source code tree.