FFMPEG-FILTERS(1) | FFMPEG-FILTERS(1) |
ffmpeg-filters - FFmpeg filters
This document describes filters, sources, and sinks provided by the libavfilter library.
Filtering in FFmpeg is enabled through the libavfilter library.
In libavfilter, a filter can have multiple inputs and multiple outputs. To illustrate the sorts of things that are possible, we consider the following filtergraph.
[main] input --> split ---------------------> overlay --> output | ^ |[tmp] [flip]| +-----> crop --> vflip -------+
This filtergraph splits the input stream in two streams, then sends one stream through the crop filter and the vflip filter, before merging it back with the other stream by overlaying it on top. You can use the following command to achieve this:
ffmpeg -i INPUT -vf "split [main][tmp]; [tmp] crop=iw:ih/2:0:0, vflip [flip]; [main][flip] overlay=0:H/2" OUTPUT
The result will be that the top half of the video is mirrored onto the bottom half of the output video.
Filters in the same linear chain are separated by commas, and distinct linear chains of filters are separated by semicolons. In our example, crop,vflip are in one linear chain, split and overlay are separately in another. The points where the linear chains join are labelled by names enclosed in square brackets. In the example, the split filter generates two outputs that are associated to the labels [main] and [tmp].
The stream sent to the second output of split, labelled as [tmp], is processed through the crop filter, which crops away the lower half part of the video, and then vertically flipped. The overlay filter takes in input the first unchanged output of the split filter (which was labelled as [main]), and overlay on its lower half the output generated by the crop,vflip filterchain.
Some filters take in input a list of parameters: they are specified after the filter name and an equal sign, and are separated from each other by a colon.
There exist so-called source filters that do not have an audio/video input, and sink filters that will not have audio/video output.
The graph2dot program included in the FFmpeg tools directory can be used to parse a filtergraph description and issue a corresponding textual representation in the dot language.
Invoke the command:
graph2dot -h
to see how to use graph2dot.
You can then pass the dot description to the dot program (from the graphviz suite of programs) and obtain a graphical representation of the filtergraph.
For example the sequence of commands:
echo <GRAPH_DESCRIPTION> | \ tools/graph2dot -o graph.tmp && \ dot -Tpng graph.tmp -o graph.png && \ display graph.png
can be used to create and display an image representing the graph described by the GRAPH_DESCRIPTION string. Note that this string must be a complete self-contained graph, with its inputs and outputs explicitly defined. For example if your command line is of the form:
ffmpeg -i infile -vf scale=640:360 outfile
your GRAPH_DESCRIPTION string will need to be of the form:
nullsrc,scale=640:360,nullsink
you may also need to set the nullsrc parameters and add a format filter in order to simulate a specific input file.
A filtergraph is a directed graph of connected filters. It can contain cycles, and there can be multiple links between a pair of filters. Each link has one input pad on one side connecting it to one filter from which it takes its input, and one output pad on the other side connecting it to one filter accepting its output.
Each filter in a filtergraph is an instance of a filter class registered in the application, which defines the features and the number of input and output pads of the filter.
A filter with no input pads is called a "source", and a filter with no output pads is called a "sink".
A filtergraph has a textual representation, which is recognized by the -filter/-vf/-af and -filter_complex options in ffmpeg and -vf/-af in ffplay, and by the "avfilter_graph_parse_ptr()" function defined in libavfilter/avfilter.h.
A filterchain consists of a sequence of connected filters, each one connected to the previous one in the sequence. A filterchain is represented by a list of ","-separated filter descriptions.
A filtergraph consists of a sequence of filterchains. A sequence of filterchains is represented by a list of ";"-separated filterchain descriptions.
A filter is represented by a string of the form: [in_link_1]...[in_link_N]filter_name@id=arguments[out_link_1]...[out_link_M]
filter_name is the name of the filter class of which the described filter is an instance of, and has to be the name of one of the filter classes registered in the program optionally followed by "@id". The name of the filter class is optionally followed by a string "=arguments".
arguments is a string which contains the parameters used to initialize the filter instance. It may have one of two forms:
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 one to name a link and associate it to a filter output or input pad. The preceding labels in_link_1 ... in_link_N, are associated to the filter input pads, the following labels out_link_1 ... out_link_M, are associated to the output pads.
When two link labels with the same name are found in the filtergraph, a link between the corresponding input and output pad is created.
If an output pad is not labelled, it is linked by default to the first unlabelled input pad of the next filter in the filterchain. For example in the filterchain
nullsrc, split[L1], [L2]overlay, nullsink
the split filter instance has two output pads, and the overlay filter instance two input pads. The first output pad of split is labelled "L1", the first input pad of overlay is labelled "L2", and the second output pad of split is linked to the second input pad of overlay, which are both unlabelled.
In a filter description, if the input label of the first filter is not specified, "in" is assumed; if the output label of the last filter is not specified, "out" is assumed.
In a complete filterchain all the unlabelled filter input and output pads must be connected. A filtergraph is considered valid if all the filter input and output pads of all the filterchains are connected.
Libavfilter will automatically insert scale filters where format conversion is required. It is possible to specify swscale flags for those automatically inserted scalers by prepending "sws_flags=flags;" to the filtergraph description.
Here is a BNF description of the filtergraph syntax:
<NAME> ::= sequence of alphanumeric characters and '_' <FILTER_NAME> ::= <NAME>["@"<NAME>] <LINKLABEL> ::= "[" <NAME> "]" <LINKLABELS> ::= <LINKLABEL> [<LINKLABELS>] <FILTER_ARGUMENTS> ::= sequence of chars (possibly quoted) <FILTER> ::= [<LINKLABELS>] <FILTER_NAME> ["=" <FILTER_ARGUMENTS>] [<LINKLABELS>] <FILTERCHAIN> ::= <FILTER> [,<FILTERCHAIN>] <FILTERGRAPH> ::= [sws_flags=<flags>;] <FILTERCHAIN> [;<FILTERGRAPH>]
Filtergraph description composition entails several levels of escaping. See the "Quoting and escaping" section in the ffmpeg-utils(1) manual for more information about the employed escaping procedure.
A first level escaping affects the content of each filter option value, which may contain the special character ":" used to separate values, or one of the escaping characters "\'".
A second level escaping affects the whole filter description, which may contain the escaping characters "\'" or the special characters "[],;" used by the filtergraph description.
Finally, when you specify a filtergraph on a shell commandline, you need to perform a third level escaping for the shell special characters contained within it.
For example, consider the following string to be embedded in the drawtext filter description text value:
this is a 'string': may contain one, or more, special characters
This string contains the "'" special escaping character, and the ":" special character, so it needs to be escaped in this way:
text=this is a \'string\'\: may contain one, or more, special characters
A second level of escaping is required when embedding the filter description in a filtergraph description, in order to escape all the filtergraph special characters. Thus the example above becomes:
drawtext=text=this is a \\\'string\\\'\\: may contain one\, or more\, special characters
(note that in addition to the "\'" escaping special characters, also "," needs to be escaped).
Finally an additional level of escaping is needed when writing the filtergraph description in a shell command, which depends on the escaping rules of the adopted shell. For example, assuming that "\" is special and needs to be escaped with another "\", the previous string will finally result in:
-vf "drawtext=text=this is a \\\\\\'string\\\\\\'\\\\: may contain one\\, or more\\, special characters"
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:
Additionally, these filters support an enable command that can be used to re-define the expression.
Like any other filtering option, the enable option follows the same rules.
For example, to enable a blur filter (smartblur) from 10 seconds to 3 minutes, and a curves filter starting at 3 seconds:
smartblur = enable='between(t,10,3*60)', curves = enable='gte(t,3)' : preset=cross_process
See "ffmpeg -filters" to view which filters have timeline support.
Some options can be changed during the operation of the filter using a command. These options are marked 'T' on the output of ffmpeg -h filter=<name of filter>. The name of the command is the name of the option and the argument is the new value.
Some filters with several inputs support a common set of options. These options can only be set by name, not with the short notation.
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.
A compressor is mainly used to reduce the dynamic range of a signal. Especially modern music is mostly compressed at a high ratio to improve the overall loudness. It's done to get the highest attention of a listener, "fatten" the sound and bring more "power" to the track. If a signal is compressed too much it may sound dull or "dead" afterwards or it may start to "pump" (which could be a powerful effect but can also destroy a track completely). The right compression is the key to reach a professional sound and is the high art of mixing and mastering. Because of its complex settings it may take a long time to get the right feeling for this kind of effect.
Compression is done by detecting the volume above a chosen level "threshold" and dividing it by the factor set with "ratio". So if you set the threshold to -12dB and your signal reaches -6dB a ratio of 2:1 will result in a signal at -9dB. Because an exact manipulation of the signal would cause distortion of the waveform the reduction can be levelled over the time. This is done by setting "Attack" and "Release". "attack" determines how long the signal has to rise above the threshold before any reduction will occur and "release" sets the time the signal has to fall below the threshold to reduce the reduction again. Shorter signals than the chosen attack time will be left untouched. The overall reduction of the signal can be made up afterwards with the "makeup" setting. So compressing the peaks of a signal about 6dB and raising the makeup to this level results in a signal twice as loud than the source. To gain a softer entry in the compression the "knee" flattens the hard edge at the threshold in the range of the chosen decibels.
The filter accepts the following options:
Commands
This filter supports the all above options as commands.
Simple audio dynamic range compression/expansion filter.
The filter accepts the following options:
Copy the input audio source unchanged to the output. This is mainly useful for testing purposes.
Apply cross fade from one input audio stream to another input audio stream. The cross fade is applied for specified duration near the end of first stream.
The filter accepts the following options:
For description of available curve types see afade filter description.
Examples
ffmpeg -i first.flac -i second.flac -filter_complex acrossfade=d=10:c1=exp:c2=exp output.flac
ffmpeg -i first.flac -i second.flac -filter_complex acrossfade=d=10:o=0:c1=exp:c2=exp output.flac
Split audio stream into several bands.
This filter splits audio stream into two or more frequency ranges. Summing all streams back will give flat output.
The filter accepts the following options:
Reduce audio bit resolution.
This filter is bit crusher with enhanced functionality. A bit crusher is used to audibly reduce number of bits an audio signal is sampled with. This doesn't change the bit depth at all, it just produces the effect. Material reduced in bit depth sounds more harsh and "digital". This filter is able to even round to continuous values instead of discrete bit depths. Additionally it has a D/C offset which results in different crushing of the lower and the upper half of the signal. An Anti-Aliasing setting is able to produce "softer" crushing sounds.
Another feature of this filter is the logarithmic mode. This setting switches from linear distances between bits to logarithmic ones. The result is a much more "natural" sounding crusher which doesn't gate low signals for example. The human ear has a logarithmic perception, so this kind of crushing is much more pleasant. Logarithmic crushing is also able to get anti-aliased.
The filter accepts the following options:
Delay audio filtering until a given wallclock timestamp. See the cue filter.
Remove impulsive noise from input audio.
Samples detected as impulsive noise are replaced by interpolated samples using autoregressive modelling.
It accepts the following values:
Default value is "a".
Remove clipped samples from input audio.
Samples detected as clipped are replaced by interpolated samples using autoregressive modelling.
It accepts the following values:
Default value is "a".
Delay one or more audio channels.
Samples in delayed channel are filled with silence.
The filter accepts the following option:
Examples
adelay=1500|0|500
adelay=0|500S|700S
adelay=delays=64S:all=1
Compute derivative/integral of audio stream.
Applying both filters one after another produces original audio.
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.
Examples
aecho=0.8:0.88:60:0.4
aecho=0.8:0.88:6:0.4
aecho=0.8:0.9:1000:0.3
aecho=0.8:0.9:1000|1800:0.3|0.25
Audio emphasis filter creates or restores material directly taken from LPs or emphased CDs with different filter curves. E.g. to store music on vinyl the signal has to be altered by a filter first to even out the disadvantages of this recording medium. Once the material is played back the inverse filter has to be applied to restore the distortion of the frequency response.
The filter accepts the following options:
Modify an audio signal according to the specified expressions.
This filter accepts one or more expressions (one for each channel), which are evaluated and used to modify a corresponding audio signal.
It accepts the following parameters:
Each expression in exprs can contain the following constants and functions:
Note: this filter is slow. For faster processing you should use a dedicated filter.
Examples
Apply fade-in/out effect to input audio.
A description of the accepted parameters follows.
It accepts the following values:
Examples
afade=t=in:ss=0:d=15
afade=t=out:st=875:d=25
Denoise audio samples with FFT.
A description of the accepted parameters follows.
It accepts the following values:
It accepts the following values:
Commands
This filter supports the following commands:
Apply arbitrary expressions to samples in frequency domain.
Each expression in real and imag can contain the following constants and functions:
Examples
afftfilt="'real=re * (1-clip((b/nb)*b,0,1))':imag='im * (1-clip((b/nb)*b,0,1))'"
afftfilt="real='hypot(re,im)*sin(0)':imag='hypot(re,im)*cos(0)':win_size=512:overlap=0.75"
afftfilt="real='hypot(re,im)*cos((random(0)*2-1)*2*3.14)':imag='hypot(re,im)*sin((random(1)*2-1)*2*3.14)':win_size=128:overlap=0.8"
Apply an arbitrary Finite Impulse Response filter.
This filter is designed for applying long FIR filters, up to 60 seconds long.
It can be used as component for digital crossover filters, room equalization, cross talk cancellation, wavefield synthesis, auralization, ambiophonics, ambisonics and spatialization.
This filter uses the streams higher than first one as FIR coefficients. If the non-first stream holds a single channel, it will be used for all input channels in the first stream, otherwise the number of channels in the non-first stream must be same as the number of channels in the first stream.
It accepts the following parameters:
Set which approach to use for auto gain measurement.
Examples
ffmpeg -i input.wav -i middle_tunnel_1way_mono.wav -lavfi afir output.wav
Set output format constraints for the input audio. The framework will negotiate the most appropriate format to minimize conversions.
It accepts the following parameters:
See the Channel Layout section in the ffmpeg-utils(1) manual for the required syntax.
If a parameter is omitted, all values are allowed.
Force the output to either unsigned 8-bit or signed 16-bit stereo
aformat=sample_fmts=u8|s16:channel_layouts=stereo
A gate is mainly used to reduce lower parts of a signal. This kind of signal processing reduces disturbing noise between useful signals.
Gating is done by detecting the volume below a chosen level threshold and dividing it by the factor set with ratio. The bottom of the noise floor is set via range. Because an exact manipulation of the signal would cause distortion of the waveform the reduction can be levelled over time. This is done by setting attack and release.
attack determines how long the signal has to fall below the threshold before any reduction will occur and release sets the time the signal has to rise above the threshold to reduce the reduction again. Shorter signals than the chosen attack time will be left untouched.
Apply an arbitrary Infinite Impulse Response filter.
It accepts the following parameters:
Coefficients in "tf" format are separated by spaces and are in ascending order.
Coefficients in "zp" format are separated by spaces and order of coefficients doesn't matter. Coefficients in "zp" format are complex numbers with i imaginary unit.
Different coefficients and gains can be provided for every channel, in such case use '|' to separate coefficients or gains. Last provided coefficients will be used for all remaining channels.
Examples
aiir=k=1:z=7.957584807809675810E-1 -2.575128568908332300 3.674839853930788710 -2.57512875289799137 7.957586296317130880E-1:p=1 -2.86950072432325953 3.63022088054647218 -2.28075678147272232 6.361362326477423500E-1:f=tf:r=d
aiir=k=0.79575848078096756:z=0.80918701+0.58773007i 0.80918701-0.58773007i 0.80884700+0.58784055i 0.80884700-0.58784055i:p=0.63892345+0.59951235i 0.63892345-0.59951235i 0.79582691+0.44198673i 0.79582691-0.44198673i:f=zp:r=s
The limiter prevents an input signal from rising over a desired threshold. This limiter uses lookahead technology to prevent your signal from distorting. It means that there is a small delay after the signal is processed. Keep in mind that the delay it produces is the attack time you set.
The filter accepts the following options:
Depending on picked setting it is recommended to upsample input 2x or 4x times with aresample before applying this filter.
Apply a two-pole all-pass filter with central frequency (in Hz) frequency, and filter-width width. An all-pass filter changes the audio's frequency to phase relationship without changing its frequency to amplitude relationship.
The filter accepts the following options:
Commands
This filter supports the following commands:
Loop audio samples.
The filter accepts the following options:
Merge two or more audio streams into a single multi-channel stream.
The filter accepts the following options:
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.
Examples
amovie=left.wav [l] ; amovie=right.mp3 [r] ; [l] [r] amerge
ffmpeg -i input.mkv -filter_complex "[0:1][0:2][0:3][0:4][0:5][0:6] amerge=inputs=6" -c:a pcm_s16le output.mkv
Mixes multiple audio inputs into a single output.
Note that this filter only supports float samples (the amerge and pan audio filters support many formats). If the amix input has integer samples then aresample will be automatically inserted to perform the conversion to float samples.
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.
It accepts the following parameters:
Commands
This filter supports the following commands:
Multiply first audio stream with second audio stream and store result in output audio stream. Multiplication is done by multiplying each sample from first stream with sample at same position from second stream.
With this element-wise multiplication one can create amplitude fades and amplitude modulations.
High-order parametric multiband equalizer for each channel.
It accepts the following parameters:
Examples
anequalizer=c0 f=200 w=100 g=-10 t=1|c1 f=200 w=100 g=-10 t=1
Commands
This filter supports the following commands:
fN is existing filter number, starting from 0, if no such filter is available error is returned. freq set new frequency parameter. width set new width parameter in herz. gain set new gain parameter in dB.
Full filter invocation with asendcmd may look like this: asendcmd=c='4.0 anequalizer change 0|f=200|w=50|g=1',anequalizer=...
Reduce broadband noise in audio samples using Non-Local Means algorithm.
Each sample is adjusted by looking for other samples with similar contexts. This context similarity is defined by comparing their surrounding patches of size p. Patches are searched in an area of r around the sample.
The filter accepts the following options:
It accepts the following values:
Commands
This filter supports the following commands:
Apply Normalized Least-Mean-Squares algorithm to the first audio stream using the second audio stream.
This adaptive filter is used to mimic a desired filter by finding the filter coefficients that relate to producing the least mean square of the error signal (difference between the desired, 2nd input audio stream and the actual signal, the 1st input audio stream).
A description of the accepted options follows.
Examples
asplit[a][b],[a]adelay=32S|32S[a],[b][a]anlms=order=128:leakage=0.0005:mu=.5:out_mode=o
Commands
This filter supports the same commands as options, excluding option "order".
Pass the audio source unchanged to the output.
Pad the end of an audio stream with silence.
This can be used together with ffmpeg -shortest to extend audio streams to the same length as the video stream.
A description of the accepted options follows.
If neither the pad_len nor the whole_len nor pad_dur nor whole_dur option is set, the filter will add silence to the end of the input stream indefinitely.
Examples
apad=pad_len=1024
apad=whole_len=10000
ffmpeg -i VIDEO -i AUDIO -filter_complex "[1:0]apad" -shortest OUTPUT
Add a phasing effect to the input audio.
A phaser filter creates series of peaks and troughs in the frequency spectrum. The position of the peaks and troughs are modulated so that they vary over time, creating a sweeping effect.
A description of the accepted parameters follows.
Audio pulsator is something between an autopanner and a tremolo. But it can produce funny stereo effects as well. Pulsator changes the volume of the left and right channel based on a LFO (low frequency oscillator) with different waveforms and shifted phases. This filter have the ability to define an offset between left and right channel. An offset of 0 means that both LFO shapes match each other. The left and right channel are altered equally - a conventional tremolo. An offset of 50% means that the shape of the right channel is exactly shifted in phase (or moved backwards about half of the frequency) - pulsator acts as an autopanner. At 1 both curves match again. Every setting in between moves the phase shift gapless between all stages and produces some "bypassing" sounds with sine and triangle waveforms. The more you set the offset near 1 (starting from the 0.5) the faster the signal passes from the left to the right speaker.
The filter accepts the following options:
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 "Resampler Options" section in the ffmpeg-resampler(1) manual for the complete list of supported options.
Examples
aresample=44100
aresample=async=1000
Reverse an audio clip.
Warning: This filter requires memory to buffer the entire clip, so trimming is suggested.
Examples
atrim=end=5,areverse
Reduce noise from speech using Recurrent Neural Networks.
This filter accepts the following options:
Set the number of samples per each output audio frame.
The last output packet may contain a different number of samples, as the filter will flush all the remaining samples when the input audio signals its end.
The filter accepts the following options:
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:
Show a line containing various information for each input audio frame. The input audio is not modified.
The shown line contains a sequence of key/value pairs of the form key:value.
The following values are shown in the output:
Apply audio soft clipping.
Soft clipping is a type of distortion effect where the amplitude of a signal is saturated along a smooth curve, rather than the abrupt shape of hard-clipping.
This filter accepts the following options:
It accepts the following values:
Commands
This filter supports the all above options as commands.
Automatic Speech Recognition
This filter uses PocketSphinx for speech recognition. To enable compilation of this filter, you need to configure FFmpeg with "--enable-pocketsphinx".
It accepts the following options:
The filter exports recognized speech as the frame metadata "lavfi.asr.text".
Display time domain statistical information about the audio channels. Statistics are calculated and displayed for each audio channel and, where applicable, an overall figure is also given.
It accepts the following option:
Available keys for each channel are: DC_offset Min_level Max_level Min_difference Max_difference Mean_difference RMS_difference Peak_level RMS_peak RMS_trough Crest_factor Flat_factor Peak_count Noise_floor Noise_floor_count Bit_depth Dynamic_range Zero_crossings Zero_crossings_rate Number_of_NaNs Number_of_Infs Number_of_denormals
and for Overall: DC_offset Min_level Max_level Min_difference Max_difference Mean_difference RMS_difference Peak_level RMS_level RMS_peak RMS_trough Flat_factor Peak_count Noise_floor Noise_floor_count Bit_depth Number_of_samples Number_of_NaNs Number_of_Infs Number_of_denormals
For example full key look like this "lavfi.astats.1.DC_offset" or this "lavfi.astats.Overall.Peak_count".
For description what each key means read below.
A description of each shown parameter follows:
Boost subwoofer frequencies.
The filter accepts the following options:
Commands
This filter supports the all above options as commands.
Adjust audio tempo.
The filter accepts exactly one parameter, the audio tempo. If not specified then the filter will assume nominal 1.0 tempo. Tempo must be in the [0.5, 100.0] range.
Note that tempo greater than 2 will skip some samples rather than blend them in. If for any reason this is a concern it is always possible to daisy-chain several instances of atempo to achieve the desired product tempo.
Examples
atempo=0.8
atempo=3
atempo=sqrt(3),atempo=sqrt(3)
Commands
This filter supports the following commands:
Trim the input so that the output contains one continuous subpart of the input.
It accepts the following parameters:
start, end, and duration are expressed as time duration specifications; see the Time duration section in the ffmpeg-utils(1) manual.
Note that the first two sets of the start/end options and the duration option look at the frame timestamp, while the _sample options simply count the samples that pass through the filter. So start/end_pts and start/end_sample will give different results when the timestamps are wrong, inexact or do not start at zero. Also note that this filter does not modify the timestamps. If you wish to have the output timestamps start at zero, insert the asetpts filter after the atrim filter.
If multiple start or end options are set, this filter tries to be greedy and keep all samples that match at least one of the specified constraints. To keep only the part that matches all the constraints at once, chain multiple atrim filters.
The defaults are such that all the input is kept. So it is possible to set e.g. just the end values to keep everything before the specified time.
Examples:
ffmpeg -i INPUT -af atrim=60:120
ffmpeg -i INPUT -af atrim=end_sample=1000
Calculate normalized cross-correlation between two input audio streams.
Resulted samples are always between -1 and 1 inclusive. If result is 1 it means two input samples are highly correlated in that selected segment. Result 0 means they are not correlated at all. If result is -1 it means two input samples are out of phase, which means they cancel each other.
The filter accepts the following options:
Examples
ffmpeg -i stereo.wav -af channelsplit,axcorrelate=size=1024:algo=fast correlation.wav
Apply a two-pole Butterworth band-pass filter with central frequency frequency, and (3dB-point) band-width width. The csg option selects a constant skirt gain (peak gain = Q) instead of the default: constant 0dB peak gain. The filter roll off at 6dB per octave (20dB per decade).
The filter accepts the following options:
Commands
This filter supports the following commands:
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:
Commands
This filter supports the following commands:
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:
Commands
This filter supports the following commands:
Apply a biquad IIR filter with the given coefficients. Where b0, b1, b2 and a0, a1, a2 are the numerator and denominator coefficients respectively. and channels, c specify which channels to filter, by default all available are filtered.
Commands
This filter supports the following commands:
Bauer stereo to binaural transformation, which improves headphone listening of stereo audio records.
To enable compilation of this filter you need to configure FFmpeg with "--enable-libbs2b".
It accepts the following parameters:
Remap input channels to new locations.
It accepts the following parameters:
If no mapping is present, the filter will implicitly map input channels to output channels, preserving indices.
Examples
ffmpeg -i in.mov -filter 'channelmap=map=DL-FL|DR-FR' out.wav
will create an output WAV file tagged as stereo from the downmix channels of the input.
ffmpeg -i in.wav -filter 'channelmap=1|2|0|5|3|4:5.1' out.wav
Split each channel from an input audio stream into a separate output stream.
It accepts the following parameters:
Choosing channels not present in channel layout in the input will result in an error.
Examples
ffmpeg -i in.mp3 -filter_complex channelsplit out.mkv
will create an output Matroska file with two audio streams, one containing only the left channel and the other the right channel.
ffmpeg -i in.wav -filter_complex 'channelsplit=channel_layout=5.1[FL][FR][FC][LFE][SL][SR]' -map '[FL]' front_left.wav -map '[FR]' front_right.wav -map '[FC]' front_center.wav -map '[LFE]' lfe.wav -map '[SL]' side_left.wav -map '[SR]' side_right.wav
ffmpeg -i in.wav -filter_complex 'channelsplit=channel_layout=5.1:channels=LFE[LFE]' -map '[LFE]' lfe.wav
Add a chorus effect to the audio.
Can make a single vocal sound like a chorus, but can also be applied to instrumentation.
Chorus resembles an echo effect with a short delay, but whereas with echo the delay is constant, with chorus, it is varied using using sinusoidal or triangular modulation. The modulation depth defines the range the modulated delay is played before or after the delay. Hence the delayed sound will sound slower or faster, that is the delayed sound tuned around the original one, like in a chorus where some vocals are slightly off key.
It accepts the following parameters:
Examples
chorus=0.7:0.9:55:0.4:0.25:2
chorus=0.6:0.9:50|60:0.4|0.32:0.25|0.4:2|1.3
chorus=0.5:0.9:50|60|40:0.4|0.32|0.3:0.25|0.4|0.3:2|2.3|1.3
Compress or expand the audio's dynamic range.
It accepts the following parameters:
The input values must be in strictly increasing order but the transfer function does not have to be monotonically rising. The point "0/0" is assumed but may be overridden (by "0/out-dBn"). Typical values for the transfer function are "-70/-70|-60/-20|1/0".
Examples
compand=.3|.3:1|1:-90/-60|-60/-40|-40/-30|-20/-20:6:0:-90:0.2
Another example for audio with whisper and explosion parts:
compand=0|0:1|1:-90/-900|-70/-70|-30/-9|0/-3:6:0:0:0
compand=.1|.1:.2|.2:-900/-900|-50.1/-900|-50/-50:.01:0:-90:.1
compand=.1|.1:.1|.1:-45.1/-45.1|-45/-900|0/-900:.01:45:-90:.1
compand=points=-80/-80|-6/-6|0/-3.8|20/3.5
compand=points=-80/-80|-9/-9|0/-5.3|20/2.9
compand=points=-80/-80|-12/-12|0/-6.8|20/1.9
compand=points=-80/-80|-18/-18|0/-9.8|20/0.7
compand=points=-80/-80|-15/-15|0/-10.8|20/-5.2
compand=points=-80/-105|-62/-80|-15.4/-15.4|0/-12|20/-7.6
compand=attacks=0:points=-80/-169|-54/-80|-49.5/-64.6|-41.1/-41.1|-25.8/-15|-10.8/-4.5|0/0|20/8.3
compand=attacks=0:points=-80/-80|-6/-6|20/-6
compand=attacks=0:points=-80/-80|-12/-12|20/-12
compand=attacks=0:points=-80/-115|-35.1/-80|-35/-35|20/20
compand=attacks=0:points=-80/-80|-12.4/-12.4|-6/-8|0/-6.8|20/-2.8
Compensation Delay Line is a metric based delay to compensate differing positions of microphones or speakers.
For example, you have recorded guitar with two microphones placed in different locations. Because the front of sound wave has fixed speed in normal conditions, the phasing of microphones can vary and depends on their location and interposition. The best sound mix can be achieved when these microphones are in phase (synchronized). Note that a distance of ~30 cm between microphones makes one microphone capture the signal in antiphase to the other microphone. That makes the final mix sound moody. This filter helps to solve phasing problems by adding different delays to each microphone track and make them synchronized.
The best result can be reached when you take one track as base and synchronize other tracks one by one with it. Remember that synchronization/delay tolerance depends on sample rate, too. Higher sample rates will give more tolerance.
The filter accepts the following parameters:
Apply headphone crossfeed filter.
Crossfeed is the process of blending the left and right channels of stereo audio recording. It is mainly used to reduce extreme stereo separation of low frequencies.
The intent is to produce more speaker like sound to the listener.
The filter accepts the following options:
Commands
This filter supports the all above options as commands.
Simple algorithm to expand audio dynamic range.
The filter accepts the following options:
Commands
This filter supports the all above options as commands.
Apply a DC shift to the audio.
This can be useful to remove a DC offset (caused perhaps by a hardware problem in the recording chain) from the audio. The effect of a DC offset is reduced headroom and hence volume. The astats filter can be used to determine if a signal has a DC offset.
Apply de-essing to the audio samples.
It accepts the following values:
Measure audio dynamic range.
DR values of 14 and higher is found in very dynamic material. DR of 8 to 13 is found in transition material. And anything less that 8 have very poor dynamics and is very compressed.
The filter accepts the following options:
Dynamic Audio Normalizer.
This filter applies a certain amount of gain to the input audio in order to bring its peak magnitude to a target level (e.g. 0 dBFS). However, in contrast to more "simple" normalization algorithms, the Dynamic Audio Normalizer *dynamically* re-adjusts the gain factor to the input audio. This allows for applying extra gain to the "quiet" sections of the audio while avoiding distortions or clipping the "loud" sections. In other words: The Dynamic Audio Normalizer will "even out" the volume of quiet and loud sections, in the sense that the volume of each section is brought to the same target level. Note, however, that the Dynamic Audio Normalizer achieves this goal *without* applying "dynamic range compressing". It will retain 100% of the dynamic range *within* each section of the audio file.
Commands
This filter supports the all above options as commands.
Make audio easier to listen to on headphones.
This filter adds `cues' to 44.1kHz stereo (i.e. audio CD format) audio so that when listened to on headphones the stereo image is moved from inside your head (standard for headphones) to outside and in front of the listener (standard for speakers).
Ported from SoX.
Apply a two-pole peaking equalisation (EQ) filter. With this filter, the signal-level at and around a selected frequency can be increased or decreased, whilst (unlike bandpass and bandreject filters) that at all other frequencies is unchanged.
In order to produce complex equalisation curves, this filter can be given several times, each with a different central frequency.
The filter accepts the following options:
Examples
equalizer=f=1000:t=h:width=200:g=-10
equalizer=f=1000:t=q:w=1:g=2,equalizer=f=100:t=q:w=2:g=-5
Commands
This filter supports the following commands:
Linearly increases the difference between left and right channels which adds some sort of "live" effect to playback.
The filter accepts the following options:
Commands
This filter supports the all above options as commands.
Apply FIR Equalization using arbitrary frequency response.
The filter accepts the following option:
and functions:
This option is also available as command. Default is gain_interpolate(f).
This option is also available as command.
Examples
firequalizer=gain='if(lt(f,1000), 0, -INF)'
firequalizer=gain_entry='entry(1000,0); entry(1001, -INF)'
firequalizer=gain_entry='entry(100,0); entry(400, -4); entry(1000, -6); entry(2000, 0)'
firequalizer=delay=0.1:fixed=on:zero_phase=on
firequalizer=gain='if(eq(chid,1), gain_interpolate(f), if(eq(chid,2), gain_interpolate(1e6+f), 0))' :gain_entry='entry(1000, 0); entry(1001,-INF); entry(1e6+1000,0)':multi=on
Apply a flanging effect to the audio.
The filter accepts the following options:
Apply Haas effect to audio.
Note that this makes most sense to apply on mono signals. With this filter applied to mono signals it give some directionality and stretches its stereo image.
The filter accepts the following options:
Decodes High Definition Compatible Digital (HDCD) data. A 16-bit PCM stream with embedded HDCD codes is expanded into a 20-bit PCM stream.
The filter supports the Peak Extend and Low-level Gain Adjustment features of HDCD, and detects the Transient Filter flag.
ffmpeg -i HDCD16.flac -af hdcd OUT24.flac
When using the filter with wav, note the default encoding for wav is 16-bit, so the resulting 20-bit stream will be truncated back to 16-bit. Use something like -acodec pcm_s24le after the filter to get 24-bit PCM output.
ffmpeg -i HDCD16.wav -af hdcd OUT16.wav ffmpeg -i HDCD16.wav -af hdcd -c:a pcm_s24le OUT24.wav
The filter accepts the following options:
"analyze_mode=pe:force_pe=true" can be used to see all samples above the PE level.
Modes are:
Apply head-related transfer functions (HRTFs) to create virtual loudspeakers around the user for binaural listening via headphones. The HRIRs are provided via additional streams, for each channel one stereo input stream is needed.
The filter accepts the following options:
Examples
ffmpeg -i input.wav -filter_complex "amovie=azi_270_ele_0_DFC.wav[sr];amovie=azi_90_ele_0_DFC.wav[sl];amovie=azi_225_ele_0_DFC.wav[br];amovie=azi_135_ele_0_DFC.wav[bl];amovie=azi_0_ele_0_DFC.wav,asplit[fc][lfe];amovie=azi_35_ele_0_DFC.wav[fl];amovie=azi_325_ele_0_DFC.wav[fr];[0:a][fl][fr][fc][lfe][bl][br][sl][sr]headphone=FL|FR|FC|LFE|BL|BR|SL|SR" output.wav
ffmpeg -i input.wav -filter_complex "amovie=minp.wav[hrirs];[0:a][hrirs]headphone=map=FL|FR|FC|LFE|BL|BR|SL|SR:hrir=multich" output.wav
Apply a high-pass filter with 3dB point frequency. The filter can be either single-pole, or double-pole (the default). The filter roll off at 6dB per pole per octave (20dB per pole per decade).
The filter accepts the following options:
Commands
This filter supports the following commands:
Join multiple input streams into one multi-channel stream.
It accepts the following parameters:
The filter will attempt to guess the mappings when they are not specified explicitly. It does so by first trying to find an unused matching input channel and if that fails it picks the first unused input channel.
Join 3 inputs (with properly set channel layouts):
ffmpeg -i INPUT1 -i INPUT2 -i INPUT3 -filter_complex join=inputs=3 OUTPUT
Build a 5.1 output from 6 single-channel streams:
ffmpeg -i fl -i fr -i fc -i sl -i sr -i lfe -filter_complex 'join=inputs=6:channel_layout=5.1:map=0.0-FL|1.0-FR|2.0-FC|3.0-SL|4.0-SR|5.0-LFE' out
Load a LADSPA (Linux Audio Developer's Simple Plugin API) plugin.
To enable compilation of this filter you need to configure FFmpeg with "--enable-ladspa".
Examples
ladspa=file=amp
ladspa=f=vcf:p=vcf_notch:c=help
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
ladspa=file=caps:Click:c=c1=20'
ladspa=caps:Eq10X2:c=c0=-48|c9=-24|c3=12|c4=2
ladspa=fast_lookahead_limiter_1913:fastLookaheadLimiter:20|0|2
ladspa=mbeq_1197:mbeq:-24|-24|-24|0|0|0|0|0|0|0|0|0|0|0|0
ladspa=caps:Narrower
ladspa=caps:White:.2
ladspa=caps:Fractal:c=c1=1
ladspa=vlevel-ladspa:vlevel_mono
Commands
This filter supports the following commands:
If the specified value is not valid, it is ignored and prior one is kept.
EBU R128 loudness normalization. Includes both dynamic and linear normalization modes. Support for both single pass (livestreams, files) and double pass (files) modes. This algorithm can target IL, LRA, and maximum true peak. In dynamic mode, to accurately detect true peaks, the audio stream will be upsampled to 192 kHz. Use the "-ar" option or "aresample" filter to explicitly set an output sample rate.
The filter accepts the following options:
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:
Examples
lowpass=c=LFE
Commands
This filter supports the following commands:
Load a LV2 (LADSPA Version 2) plugin.
To enable compilation of this filter you need to configure FFmpeg with "--enable-lv2".
Examples
lv2=p=http\\\\://calf.sourceforge.net/plugins/BassEnhancer:c=amount=2
lv2=p=http\\\\://calf.sourceforge.net/plugins/Vinyl:c=drone=0.2|aging=0.5
lv2=p=http\\\\://www.openavproductions.com/artyfx#bitta:c=crush=0.3
Multiband Compress or expand the audio's dynamic range.
The input audio is divided into bands using 4th order Linkwitz-Riley IIRs. This is akin to the crossover of a loudspeaker, and results in flat frequency response when absent compander action.
It accepts the following parameters:
Mix channels with specific gain levels. The filter accepts the output channel layout followed by a set of channels definitions.
This filter is also designed to efficiently remap the channels of an audio stream.
The filter accepts parameters of the form: "l|outdef|outdef|..."
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.
Mixing examples
For example, if you want to down-mix from stereo to mono, but with a bigger factor for the left channel:
pan=1c|c0=0.9*c0+0.1*c1
A customized down-mix to stereo that works automatically for 3-, 4-, 5- and 7-channels surround:
pan=stereo| FL < FL + 0.5*FC + 0.6*BL + 0.6*SL | FR < FR + 0.5*FC + 0.6*BR + 0.6*SR
Note that ffmpeg integrates a default down-mix (and up-mix) system that should be preferred (see "-ac" option) unless you have very specific needs.
Remapping examples
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. It is not meant to be used directly.
Apply time-stretching and pitch-shifting with librubberband.
To enable compilation of this filter, you need to configure FFmpeg with "--enable-librubberband".
The filter accepts the following options:
Commands
This filter supports the following commands:
This filter acts like normal compressor but has the ability to compress detected signal using second input signal. It needs two input streams and returns one output stream. First input stream will be processed depending on second stream signal. The filtered signal then can be filtered with other filters in later stages of processing. See pan and amerge filter.
The filter accepts the following options:
Commands
This filter supports the all above options as commands.
Examples
ffmpeg -i main.flac -i sidechain.flac -filter_complex "[1:a]asplit=2[sc][mix];[0:a][sc]sidechaincompress[compr];[compr][mix]amerge"
A sidechain gate acts like a normal (wideband) gate but has the ability to filter the detected signal before sending it to the gain reduction stage. Normally a gate uses the full range signal to detect a level above the threshold. For example: If you cut all lower frequencies from your sidechain signal the gate will decrease the volume of your track only if not enough highs appear. With this technique you are able to reduce the resonation of a natural drum or remove "rumbling" of muted strokes from a heavily distorted guitar. It needs two input streams and returns one output stream. First input stream will be processed depending on second stream signal.
The filter accepts the following options:
Detect silence in an audio stream.
This filter logs a message when it detects that the input audio volume is less or equal to a noise tolerance value for a duration greater or equal to the minimum detected noise duration.
The printed times and duration are expressed in seconds. The "lavfi.silence_start" or "lavfi.silence_start.X" metadata key is set on the first frame whose timestamp equals or exceeds the detection duration and it contains the timestamp of the first frame of the silence.
The "lavfi.silence_duration" or "lavfi.silence_duration.X" and "lavfi.silence_end" or "lavfi.silence_end.X" metadata keys are set on the first frame after the silence. If mono is enabled, and each channel is evaluated separately, the ".X" suffixed keys are used, and "X" corresponds to the channel number.
The filter accepts the following options:
Examples
silencedetect=n=-50dB:d=5
ffmpeg -i silence.mp3 -af silencedetect=noise=0.0001 -f null -
Remove silence from the beginning, middle or end of the audio.
The filter accepts the following options:
Examples
silenceremove=start_periods=1:start_duration=5:start_threshold=0.02
silenceremove=stop_periods=-1:stop_duration=1:stop_threshold=-90dB
silenceremove=window=0:detection=peak:stop_mode=all:start_mode=all:stop_periods=-1:stop_threshold=0
SOFAlizer uses head-related transfer functions (HRTFs) to create virtual loudspeakers around the user for binaural listening via headphones (audio formats up to 9 channels supported). The HRTFs are stored in SOFA files (see <http://www.sofacoustics.org/> for a database). SOFAlizer is developed at the Acoustics Research Institute (ARI) of the Austrian Academy of Sciences.
To enable compilation of this filter you need to configure FFmpeg with "--enable-libmysofa".
The filter accepts the following options:
Examples
sofalizer=sofa=/path/to/ClubFritz6.sofa:type=freq:radius=1
sofalizer=sofa=/path/to/ClubFritz12.sofa:type=freq:radius=2:rotation=5
"sofalizer=sofa=/path/to/ClubFritz6.sofa:type=freq:radius=2:speakers=FL 45|FR 315|BL 135|BR 225:gain=28"
This filter has some handy utilities to manage stereo signals, for converting M/S stereo recordings to L/R signal while having control over the parameters or spreading the stereo image of master track.
The filter accepts the following options:
Can be one of the following:
Examples
stereotools=mlev=0.015625
"stereotools=mode=ms>lr"
This filter enhance the stereo effect by suppressing signal common to both channels and by delaying the signal of left into right and vice versa, thereby widening the stereo effect.
The filter accepts the following options:
Commands
This filter supports the all above options except "delay" as commands.
Apply 18 band equalizer.
The filter accepts the following options:
Apply audio surround upmix filter.
This filter allows to produce multichannel output from audio stream.
The filter accepts the following options:
See the Channel Layout section in the ffmpeg-utils(1) manual for the required syntax.
See the Channel Layout section in the ffmpeg-utils(1) manual for the required syntax.
It accepts the following values:
Default is "hann".
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:
Commands
This filter supports the following commands:
Sinusoidal amplitude modulation.
The filter accepts the following options:
Sinusoidal phase modulation.
The filter accepts the following options:
Adjust the input audio volume.
It accepts the following parameters:
Output values are clipped to the maximum value.
The output audio volume is given by the relation:
<output_volume> = <volume> * <input_volume>
The default value for volume is "1.0".
It determines which input sample formats will be allowed, which affects the precision of the volume scaling.
Default value for replaygain_preamp is 0.0.
Default value for replaygain_noclip is 1.
It accepts the following values:
Default value is once.
The volume expression can contain the following parameters.
Note that when eval is set to once only the sample_rate and tb variables are available, all other variables will evaluate to NAN.
Commands
This filter supports the following commands:
If the specified expression is not valid, it is kept at its current value.
Examples
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.
Examples
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 parameters:
Examples
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:
If not specified, or the expressed duration is negative, the audio is supposed to be generated forever.
Each expression in exprs can contain the following constants:
Examples
aevalsrc=0
aevalsrc="sin(440*2*PI*t):s=8000"
aevalsrc="sin(420*2*PI*t)|cos(430*2*PI*t):c=FC|BC"
aevalsrc="-2+random(0)"
aevalsrc="sin(10*2*PI*t)*sin(880*2*PI*t)"
aevalsrc="0.1*sin(2*PI*(360-2.5/2)*t) | 0.1*sin(2*PI*(360+2.5/2)*t)"
Generate a FIR coefficients using frequency sampling method.
The resulting stream can be used with afir filter for filtering the audio signal.
The filter accepts the following options:
The null audio source, return unprocessed audio frames. It is mainly useful as a template and to be employed in analysis / debugging tools, or as the source for filters which ignore the input data (for example the sox synth filter).
This source accepts the following options:
Check the channel_layout_map definition in libavutil/channel_layout.c for the mapping between strings and channel layout values.
Examples
anullsrc=r=48000:cl=4
anullsrc=r=48000:cl=mono
All the parameters need to be explicitly defined.
Synthesize a voice utterance using the libflite library.
To enable compilation of this filter you need to configure FFmpeg with "--enable-libflite".
Note that versions of the flite library prior to 2.0 are not thread-safe.
The filter accepts the following options:
Examples
flite=textfile=speech.txt
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
ffplay -f lavfi flite=text='No more be grieved for which that thou hast done.'
For more information about libflite, check: <http://www.festvox.org/flite/>
Generate a noise audio signal.
The filter accepts the following options:
Examples
anoisesrc=d=60:c=pink:r=44100:a=0.5
Generate odd-tap Hilbert transform FIR coefficients.
The resulting stream can be used with afir filter for phase-shifting the signal by 90 degrees.
This is used in many matrix coding schemes and for analytic signal generation. The process is often written as a multiplication by i (or j), the imaginary unit.
The filter accepts the following options:
Generate a sinc kaiser-windowed low-pass, high-pass, band-pass, or band-reject FIR coefficients.
The resulting stream can be used with afir filter for filtering the audio signal.
The filter accepts the following options:
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:
The expression can contain the following constants:
Default is 1024.
Examples
sine
sine=220:4:d=5 sine=f=220:b=4:d=5 sine=frequency=220:beep_factor=4:duration=5
sine=1000:samples_per_frame='st(0,mod(n,5)); 1602-not(not(eq(ld(0),1)+eq(ld(0),3)))'
Below is a description of the currently available audio sinks.
Buffer audio frames, and make them available to the end of filter chain.
This sink is mainly intended for programmatic use, in particular through the interface defined in libavfilter/buffersink.h or the options system.
It accepts a pointer to an AVABufferSinkContext structure, which defines the incoming buffers' formats, to be passed as the opaque parameter to "avfilter_init_filter" for initialization.
Null audio sink; do absolutely nothing with the input audio. It is mainly useful as a template and for use in analysis / debugging tools.
When you configure your FFmpeg build, you can disable any of the existing filters using "--disable-filters". The configure output will show the video filters included in your build.
Below is a description of the currently available video filters.
Mark a region of interest in a video frame.
The frame data is passed through unchanged, but metadata is attached to the frame indicating regions of interest which can affect the behaviour of later encoding. Multiple regions can be marked by applying the filter multiple times.
The parameters x, y, w and h are expressions, and may contain the following variables:
This must be a real value in the range -1 to +1. A value of zero indicates no quality change. A negative value asks for better quality (less quantisation), while a positive value asks for worse quality (greater quantisation).
The range is calibrated so that the extreme values indicate the largest possible offset - if the rest of the frame is encoded with the worst possible quality, an offset of -1 indicates that this region should be encoded with the best possible quality anyway. Intermediate values are then interpolated in some codec-dependent way.
For example, in 10-bit H.264 the quantisation parameter varies between -12 and 51. A typical qoffset value of -1/10 therefore indicates that this region should be encoded with a QP around one-tenth of the full range better than the rest of the frame. So, if most of the frame were to be encoded with a QP of around 30, this region would get a QP of around 24 (an offset of approximately -1/10 * (51 - -12) = -6.3). An extreme value of -1 would indicate that this region should be encoded with the best possible quality regardless of the treatment of the rest of the frame - that is, should be encoded at a QP of -12.
Examples
addroi=iw/4:ih/4:iw/2:ih/2:-1/10
addroi=0:0:100:ih:+1/5
Extract the alpha component from the input as a grayscale video. This is especially useful with the alphamerge filter.
Add or replace the alpha component of the primary input with the grayscale value of a second input. This is intended for use with alphaextract to allow the transmission or storage of frame sequences that have alpha in a format that doesn't support an alpha channel.
For example, to reconstruct full frames from a normal YUV-encoded video and a separate video created with alphaextract, you might use:
movie=in_alpha.mkv [alpha]; [in][alpha] alphamerge [out]
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.
Amplify differences between current pixel and pixels of adjacent frames in same pixel location.
This filter accepts the following options:
Commands
This filter supports the following commands that corresponds to option of same name:
Same as the subtitles filter, except that it doesn't require libavcodec and libavformat to work. On the other hand, it is limited to ASS (Advanced Substation Alpha) subtitles files.
This filter accepts the following option in addition to the common options from the subtitles filter:
Available values are:
The default is "auto".
Apply an Adaptive Temporal Averaging Denoiser to the video input.
The filter accepts the following options:
Threshold A is designed to react on abrupt changes in the input signal and threshold B is designed to react on continuous changes in the input signal.
Parallel can be faster then serial, while other way around is never true. Parallel will abort early on first change being greater then thresholds, while serial will continue processing other side of frames if they are equal or bellow thresholds.
Commands
This filter supports same commands as options except option "s". The command accepts the same syntax of the corresponding option.
Apply average blur filter.
The filter accepts the following options:
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Compute 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:
Apply bilateral filter, spatial smoothing while preserving edges.
The filter accepts the following options:
Show and measure bit plane noise.
The filter accepts the following options:
Detect video intervals that are (almost) completely black. Can be useful to detect chapter transitions, commercials, or invalid recordings.
The filter outputs its detection analysis to both the log as well as frame metadata. If a black segment of at least the specified minimum duration is found, a line with the start and end timestamps as well as duration is printed to the log with level "info". In addition, a log line with level "debug" is printed per frame showing the black amount detected for that frame.
The filter also attaches metadata to the first frame of a black segment with key "lavfi.black_start" and to the first frame after the black segment ends with key "lavfi.black_end". The value is the frame's timestamp. This metadata is added regardless of the minimum duration specified.
The filter accepts the following options:
Default value is 2.0.
<nb_black_pixels> / <nb_pixels>
for which a picture is considered black. Default value is 0.98.
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.
This filter exports frame metadata "lavfi.blackframe.pblack". The value represents the percentage of pixels in the picture that are below the threshold value.
It accepts the following parameters:
Blend two video frames into each other.
The "blend" filter takes two input streams and outputs one stream, the first input is the "top" layer and second input is "bottom" layer. By default, the output terminates when the longest input terminates.
The "tblend" (time blend) filter takes two consecutive frames from one single stream, and outputs the result obtained by blending the new frame on top of the old frame.
A description of the accepted options follows.
Available values for component modes are:
The expressions can use the following variables:
The "blend" filter also supports the framesync options.
Examples
blend=all_expr='A*(if(gte(T,10),1,T/10))+B*(1-(if(gte(T,10),1,T/10)))'
blend=all_expr='A*(X/W)+B*(1-X/W)'
blend=all_expr='if(eq(mod(X,2),mod(Y,2)),A,B)'
blend=all_expr='if(gte(N*SW+X,W),A,B)'
blend=all_expr='if(gte(Y-N*SH,0),A,B)'
blend=all_expr='if(gte(T*SH*40+Y,H)*gte((T*40*SW+X)*W/H,W),A,B)'
blend=all_expr='if(gt(X,Y*(W/H)),A,B)'
tblend=all_mode=grainextract
Denoise frames using Block-Matching 3D algorithm.
The filter accepts the following options.
Examples
bm3d=sigma=3:block=4:bstep=2:group=1:estim=basic
bm3d=sigma=3:block=4:bstep=2:group=1:estim=basic:planes=1
split[a][b],[a]bm3d=sigma=3:block=4:bstep=2:group=1:estim=basic[a],[b][a]bm3d=sigma=3:block=4:bstep=2:group=16:estim=final:ref=1
split[a][b],[a]nlmeans=s=3:r=7:p=3[a],[b][a]bm3d=sigma=3:block=4:bstep=2:group=16:estim=final:ref=1
Apply a boxblur algorithm to the input video.
It accepts the following parameters:
A description of the accepted options follows.
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:
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.
Examples
boxblur=luma_radius=2:luma_power=1 boxblur=2:1
boxblur=2:1:cr=0:ar=0
boxblur=luma_radius=min(h\,w)/10:luma_power=1:chroma_radius=min(cw\,ch)/10:chroma_power=1
Deinterlace the input video ("bwdif" stands for "Bob Weaver Deinterlacing Filter").
Motion adaptive deinterlacing based on yadif with the use of w3fdif and cubic interpolation algorithms. It accepts the following parameters:
The default value is "send_field".
The default value is "auto". If the interlacing is unknown or the decoder does not export this information, top field first will be assumed.
The default value is "all".
Apply Contrast Adaptive Sharpen filter to video stream.
The filter accepts the following options:
Remove all color information for all colors except for certain one.
The filter accepts the following options:
Literal colors like "green" or "red" don't make sense with this enabled anymore. This can be used to pass exact YUV values as hexadecimal numbers.
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
YUV colorspace color/chroma keying.
The filter accepts the following options:
0.01 matches only the exact key color, while 1.0 matches everything.
0.0 makes pixels either fully transparent, or not transparent at all.
Higher values result in semi-transparent pixels, with a higher transparency the more similar the pixels color is to the key color.
Literal colors like "green" or "red" don't make sense with this enabled anymore. This can be used to pass exact YUV values as hexadecimal numbers.
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Examples
ffmpeg -i input.png -vf chromakey=green out.png
ffmpeg -f lavfi -i color=c=black:s=1280x720 -i video.mp4 -shortest -filter_complex "[1:v]chromakey=0x70de77:0.1:0.2[ckout];[0:v][ckout]overlay[out]" -map "[out]" output.mkv
Shift chroma pixels horizontally and/or vertically.
The filter accepts the following options:
Commands
This filter supports the all above options as commands.
Display CIE color diagram with pixels overlaid onto it.
The filter accepts the following options:
See "system" option for available values.
Visualize information exported by some codecs.
Some codecs can export information through frames using side-data or other means. For example, some MPEG based codecs export motion vectors through the export_mvs flag in the codec flags2 option.
The filter accepts the following option:
Available flags for mv are:
Available flags for mv_type are:
Available flags for frame_type are:
Examples
ffplay -flags2 +export_mvs input.mp4 -vf codecview=mv_type=fp
ffplay -flags2 +export_mvs input.mp4 -vf codecview=mv=pf+bf+bb
Modify intensity of primary colors (red, green and blue) of input frames.
The filter allows an input frame to be adjusted in the shadows, midtones or highlights regions for the red-cyan, green-magenta or blue-yellow balance.
A positive adjustment value shifts the balance towards the primary color, a negative value towards the complementary color.
The filter accepts the following options:
Allowed ranges for options are "[-1.0, 1.0]". Defaults are 0.
Examples
colorbalance=rs=.3
Commands
This filter supports the all above options as commands.
Adjust video input frames by re-mixing color channels.
This filter modifies a color channel by adding the values associated to the other channels of the same pixels. For example if the value to modify is red, the output value will be:
<red>=<red>*<rr> + <blue>*<rb> + <green>*<rg> + <alpha>*<ra>
The filter accepts the following options:
Allowed ranges for options are "[-2.0, 2.0]".
Examples
colorchannelmixer=.3:.4:.3:0:.3:.4:.3:0:.3:.4:.3
colorchannelmixer=.393:.769:.189:0:.349:.686:.168:0:.272:.534:.131
Commands
This filter supports the all above options as commands.
RGB colorspace color keying.
The filter accepts the following options:
0.01 matches only the exact key color, while 1.0 matches everything.
0.0 makes pixels either fully transparent, or not transparent at all.
Higher values result in semi-transparent pixels, with a higher transparency the more similar the pixels color is to the key color.
Examples
ffmpeg -i input.png -vf colorkey=green out.png
ffmpeg -i background.png -i video.mp4 -filter_complex "[1:v]colorkey=0x3BBD1E:0.3:0.2[ckout];[0:v][ckout]overlay[out]" -map "[out]" output.flv
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Remove all color information for all RGB colors except for certain one.
The filter accepts the following options:
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Adjust video input frames using levels.
The filter accepts the following options:
Input levels are used to lighten highlights (bright tones), darken shadows (dark tones), change the balance of bright and dark tones.
Output levels allows manual selection of a constrained output level range.
Examples
colorlevels=rimin=0.058:gimin=0.058:bimin=0.058
colorlevels=rimin=0.039:gimin=0.039:bimin=0.039:rimax=0.96:gimax=0.96:bimax=0.96
colorlevels=rimax=0.902:gimax=0.902:bimax=0.902
colorlevels=romin=0.5:gomin=0.5:bomin=0.5
Commands
This filter supports the all above options as commands.
Convert color matrix.
The filter accepts the following options:
The accepted values are:
For example to convert from BT.601 to SMPTE-240M, use the command:
colormatrix=bt601:smpte240m
Convert colorspace, transfer characteristics or color primaries. Input video needs to have an even size.
The filter accepts the following options:
The accepted values are:
The accepted values are:
The accepted values are:
The accepted values are:
The accepted values are:
The accepted values are:
The accepted values are:
The accepted values are:
The filter converts the transfer characteristics, color space and color primaries to the specified user values. The output value, if not specified, is set to a default value based on the "all" property. If that property is also not specified, the filter will log an error. The output color range and format default to the same value as the input color range and format. The input transfer characteristics, color space, color primaries and color range should be set on the input data. If any of these are missing, the filter will log an error and no conversion will take place.
For example to convert the input to SMPTE-240M, use the command:
colorspace=smpte240m
Apply convolution of 3x3, 5x5, 7x7 or horizontal/vertical up to 49 elements.
The filter accepts the following options:
Examples
convolution="0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0"
convolution="1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1/9:1/9:1/9:1/9"
convolution="0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:5:1:1:1:0:128:128:128"
convolution="0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:5:5:5:1:0:128:128:128"
convolution="1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:5:5:5:1:0:128:128:0"
convolution="-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2"
Apply 2D convolution of video stream in frequency domain using second stream as impulse.
The filter accepts the following options:
The "convolve" filter also supports the framesync options.
Copy the input video source unchanged to the output. This is mainly useful for testing purposes.
Video filtering on GPU using Apple's CoreImage API on OSX.
Hardware acceleration is based on an OpenGL context. Usually, this means it is processed by video hardware. However, software-based OpenGL implementations exist which means there is no guarantee for hardware processing. It depends on the respective OSX.
There are many filters and image generators provided by Apple that come with a large variety of options. The filter has to be referenced by its name along with its options.
The coreimage filter accepts the following options:
list_filters=true
It is required to specify either "default" or at least one of the filter options. All omitted options are used with their default values. The syntax of the filter string is as follows:
filter=<NAME>@<OPTION>=<VALUE>[@<OPTION>=<VALUE>][@...][#<NAME>@<OPTION>=<VALUE>[@<OPTION>=<VALUE>][@...]][#...]
output_rect=x\ y\ width\ height
If not given, the output rectangle equals the dimensions of the input image. The output rectangle is automatically cropped at the borders of the input image. Negative values are valid for each component.
output_rect=25\ 25\ 100\ 100
Several filters can be chained for successive processing without GPU-HOST transfers allowing for fast processing of complex filter chains. Currently, only filters with zero (generators) or exactly one (filters) input image and one output image are supported. Also, transition filters are not yet usable as intended.
Some filters generate output images with additional padding depending on the respective filter kernel. The padding is automatically removed to ensure the filter output has the same size as the input image.
For image generators, the size of the output image is determined by the previous output image of the filter chain or the input image of the whole filterchain, respectively. The generators do not use the pixel information of this image to generate their output. However, the generated output is blended onto this image, resulting in partial or complete coverage of the output image.
The coreimagesrc video source can be used for generating input images which are directly fed into the filter chain. By using it, providing input images by another video source or an input video is not required.
Examples
coreimage=list_filters=true
coreimage=filter=CIBoxBlur@default
coreimage=filter=CIBoxBlur@default#CIVignetteEffect@inputCenter=100\ 100@inputRadius=50
ffmpeg -f lavfi -i nullsrc=s=100x100,coreimage=filter=CIQRCodeGenerator@inputMessage=https\\\\\://FFmpeg.org/@inputCorrectionLevel=H -frames:v 1 QRCode.png
Cover a rectangular object
It accepts the following options:
It accepts the following values:
Default value is blur.
Examples
ffmpeg -i file.ts -vf find_rect=newref.pgm,cover_rect=cover.jpg:mode=cover new.mkv
Crop the input video to given dimensions.
It accepts the following parameters:
The out_w, out_h, x, y parameters are expressions containing the following constants:
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.
Examples
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)
Commands
This filter supports the following commands:
Auto-detect the crop size.
It calculates the necessary cropping parameters and prints the recommended parameters via the logging system. The detected dimensions correspond to the non-black area of the input video.
It accepts the following parameters:
This can be useful when channel logos distort the video area. 0 indicates 'never reset', and returns the largest area encountered during playback.
Delay video filtering until a given wallclock timestamp. The filter first passes on preroll amount of frames, then it buffers at most buffer amount of frames and waits for the cue. After reaching the cue it forwards the buffered frames and also any subsequent frames coming in its input.
The filter can be used synchronize the output of multiple ffmpeg processes for realtime output devices like decklink. By putting the delay in the filtering chain and pre-buffering frames the process can pass on data to output almost immediately after the target wallclock timestamp is reached.
Perfect frame accuracy cannot be guaranteed, but the result is good enough for some use cases.
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.
The filter accepts the following options:
Default is "none".
To avoid some filtergraph syntax conflicts, each key points list need to be defined using the following syntax: "x0/y0 x1/y1 x2/y2 ...".
Examples
curves=blue='0/0 0.5/0.58 1/1'
curves=r='0/0.11 .42/.51 1/0.95':g='0/0 0.50/0.48 1/1':b='0/0.22 .49/.44 1/0.8'
Here we obtain the following coordinates for each components:
curves=preset=vintage
curves=vintage
curves=psfile='MyCurvesPresets/purple.acv':green='0/0 0.45/0.53 1/1'
ffmpeg -f lavfi -i color -vf curves=cross_process:plot=/tmp/curves.plt -frames:v 1 -f null - gnuplot -p /tmp/curves.plt
Video data analysis filter.
This filter shows hexadecimal pixel values of part of video.
The filter accepts the following options:
Apply Directional blur filter.
The filter accepts the following options:
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Denoise frames using 2D DCT (frequency domain filtering).
This filter is not designed for real time.
The filter accepts the following options:
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.
If the overlapping value doesn't permit processing the whole input width or height, a warning will be displayed and according borders won't be denoised.
Default value is blocksize-1, which is the best possible setting.
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.
The default value is 3 (8x8) and can be raised to 4 for a blocksize of 16x16. Note that changing this setting has huge consequences on the speed processing. Also, a larger block size does not necessarily means a better de-noising.
Examples
Apply a denoise with a sigma of 4.5:
dctdnoiz=4.5
The same operation can be achieved using the expression system:
dctdnoiz=e='gte(c, 4.5*3)'
Violent denoise using a block size of "16x16":
dctdnoiz=15:n=4
Remove banding artifacts from input video. It works by replacing banded pixels with average value of referenced pixels.
The filter accepts the following options:
Remove blocking artifacts from input video.
The filter accepts the following options:
Examples
deblock=filter=weak:block=4
deblock=filter=strong:block=4:alpha=0.12:beta=0.07:gamma=0.06:delta=0.05
deblock=filter=strong:block=4:alpha=0.12:beta=0.07:gamma=0.06:delta=0.05:planes=1
deblock=filter=strong:block=4:alpha=0.12:beta=0.07:gamma=0.06:delta=0.05:planes=6
Drop duplicated frames at regular intervals.
The filter accepts the following options:
Apply 2D deconvolution of video stream in frequency domain using second stream as impulse.
The filter accepts the following options:
The "deconvolve" filter also supports the framesync options.
Reduce cross-luminance (dot-crawl) and cross-color (rainbows) from video.
It accepts the following options:
Apply deflate effect to the video.
This filter replaces the pixel by the local(3x3) average by taking into account only values lower than the pixel.
It accepts the following options:
Commands
This filter supports the all above options as commands.
Remove temporal frame luminance variations.
It accepts the following options:
Available values are:
Remove judder produced by partially interlaced telecined content.
Judder can be introduced, for instance, by pullup filter. If the original source was partially telecined content then the output of "pullup,dejudder" will have a variable frame rate. May change the recorded frame rate of the container. Aside from that change, this filter will not affect constant frame rate video.
The option available in this filter is:
Accepts any integer greater than 1. Useful values are:
The default is 4.
Suppress a TV station logo by a simple interpolation of the surrounding pixels. Just set a rectangle covering the logo and watch it disappear (and sometimes something even uglier appear - your mileage may vary).
It accepts the following parameters:
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.
Examples
delogo=x=0:y=0:w=100:h=77:band=10
Remove the rain in the input image/video by applying the derain methods based on convolutional neural networks. Supported models:
Training as well as model generation scripts are provided in the repository at <https://github.com/XueweiMeng/derain_filter.git>.
Native model files (.model) can be generated from TensorFlow model files (.pb) by using tools/python/convert.py
The filter accepts the following options:
Default value is derain.
Default value is native.
It can also be finished with dnn_processing filter.
Attempt to fix small changes in horizontal and/or vertical shift. This filter helps remove camera shake from hand-holding a camera, bumping a tripod, moving on a vehicle, etc.
The filter accepts the following options:
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.
Default value is mirror.
Default value is exhaustive.
Remove unwanted contamination of foreground colors, caused by reflected color of greenscreen or bluescreen.
This filter accepts the following options:
Apply an exact inverse of the telecine operation. It requires a predefined pattern specified using the pattern option which must be the same as that passed to the telecine filter.
This filter accepts the following options:
Apply dilation effect to the video.
This filter replaces the pixel by the local(3x3) maximum.
It accepts the following options:
Flags to local 3x3 coordinates maps like this:
1 2 3 4 5 6 7 8
Commands
This filter supports the all above options as commands.
Displace pixels as indicated by second and third input stream.
It takes three input streams and outputs one stream, the first input is the source, and second and third input are displacement maps.
The second input specifies how much to displace pixels along the x-axis, while the third input specifies how much to displace pixels along the y-axis. If one of displacement map streams terminates, last frame from that displacement map will be used.
Note that once generated, displacements maps can be reused over and over again.
A description of the accepted options follows.
Available values are:
Default is smear.
Examples
ffmpeg -i INPUT -f lavfi -i nullsrc=s=hd720,lutrgb=128:128:128 -f lavfi -i nullsrc=s=hd720,geq='r=128+30*sin(2*PI*X/400+T):g=128+30*sin(2*PI*X/400+T):b=128+30*sin(2*PI*X/400+T)' -lavfi '[0][1][2]displace' OUTPUT
ffmpeg -i INPUT -f lavfi -i nullsrc=hd720,geq='r=128+80*(sin(sqrt((X-W/2)*(X-W/2)+(Y-H/2)*(Y-H/2))/220*2*PI+T)):g=128+80*(sin(sqrt((X-W/2)*(X-W/2)+(Y-H/2)*(Y-H/2))/220*2*PI+T)):b=128+80*(sin(sqrt((X-W/2)*(X-W/2)+(Y-H/2)*(Y-H/2))/220*2*PI+T))' -lavfi '[1]split[x][y],[0][x][y]displace' OUTPUT
Do image processing with deep neural networks. It works together with another filter which converts the pixel format of the Frame to what the dnn network requires.
The filter accepts the following options:
Default value is native.
Native model file (.model) can be generated from TensorFlow model file (.pb) by using tools/python/convert.py
Examples
./ffmpeg -i rain.jpg -vf format=rgb24,dnn_processing=dnn_backend=tensorflow:model=can.pb:input=x:output=y derain.jpg
ffmpeg -i input.jpg -vf format=grayf32,dnn_processing=model=halve_gray_float.model:input=dnn_in:output=dnn_out:dnn_backend=native -y out.native.png
./ffmpeg -i 480p.jpg -vf format=yuv420p,scale=w=iw*2:h=ih*2,dnn_processing=dnn_backend=tensorflow:model=srcnn.pb:input=x:output=y -y srcnn.jpg
./ffmpeg -i 480p.jpg -vf format=yuv420p,dnn_processing=dnn_backend=tensorflow:model=espcn.pb:input=x:output=y -y tmp.espcn.jpg
Draw a colored box on the input image.
It accepts the following parameters:
See below for the list of accepted constants.
The parameters for x, y, w and h and t are expressions containing the following constants:
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".
Examples
drawbox
drawbox=10:20:200:60:red@0.5
The previous example can be specified as:
drawbox=x=10:y=20:w=200:h=60:color=red@0.5
drawbox=x=10:y=10:w=100:h=100:color=pink@0.5:t=fill
drawbox=x=-t:y=0.5*(ih-iw/2.4)-t:w=iw+t*2:h=iw/2.4+t*2:t=2:c=red
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Draw a graph using input video metadata.
It accepts the following parameters:
Available values for mode is:
Default is "line".
Available values for slide is:
Default is "frame".
The foreground color expressions can use the following variables:
The color is defined as 0xAABBGGRR.
Example using metadata from signalstats filter:
signalstats,drawgraph=lavfi.signalstats.YAVG:min=0:max=255
Example using metadata from ebur128 filter:
ebur128=metadata=1,adrawgraph=lavfi.r128.M:min=-120:max=5
Draw a grid on the input image.
It accepts the following parameters:
See below for the list of accepted constants.
The parameters for x, y, w and h and t are expressions containing the following constants:
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".
Examples
drawgrid=width=100:height=100:thickness=2:color=red@0.5
drawgrid=w=iw/3:h=ih/3:t=2:c=white@0.5
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Draw a text string or text from a specified file on top of a video, using the libfreetype library.
To enable compilation of this filter, you need to configure FFmpeg with "--enable-libfreetype". To enable default font fallback and the font option you need to configure FFmpeg with "--enable-libfontconfig". To enable the text_shaping option, you need to configure FFmpeg with "--enable-libfribidi".
Syntax
It accepts the following parameters:
The default value of boxcolor is "white".
The default value of bordercolor is "black".
The default value of fontcolor is "black".
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 default value of shadowcolor is "black".
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.
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:
These parameters allow the x and y expressions to refer to each other, so you can for example specify "y=x/dar".
Text expansion
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.
Sequences of the form "%{...}" are expanded. The text between the braces is a function name, possibly followed by arguments separated by ':'. If the arguments contain special characters or delimiters (':' or '}'), they should be escaped.
Note that they probably must also be escaped as the value for the text option in the filter argument string and as the filter argument in the filtergraph description, and possibly also for the shell, that makes up to four levels of escaping; using a text file avoids these problems.
The following functions are available:
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.
The first argument is the expression to be evaluated, just as for the expr function. The second argument specifies the output format. Allowed values are x, X, d and u. They are treated exactly as in the "printf" function. The third parameter is optional and sets the number of positions taken by the output. It can be used to add padding with zeros from the left.
The first argument is mandatory and specifies the metadata key.
The second argument is optional and specifies a default value, used when the metadata key is not found or empty.
Available metadata can be identified by inspecting entries starting with TAG included within each frame section printed by running "ffprobe -show_frames".
String metadata generated in filters leading to the drawtext filter are also available.
The first argument is the format of the timestamp; it defaults to "flt" for seconds as a decimal number with microsecond accuracy; "hms" stands for a formatted [-]HH:MM:SS.mmm timestamp with millisecond accuracy. "gmtime" stands for the timestamp of the frame formatted as UTC time; "localtime" stands for the timestamp of the frame formatted as local time zone time.
The second argument is an offset added to the timestamp.
If the format is set to "hms", a third argument "24HH" may be supplied to present the hour part of the formatted timestamp in 24h format (00-23).
If the format is set to "localtime" or "gmtime", a third argument may be supplied: a strftime() format string. By default, YYYY-MM-DD HH:MM:SS format will be used.
Commands
This filter supports altering parameters via commands:
Syntax for the argument is the same as for filter invocation, e.g.
fontsize=56:fontcolor=green:text='Hello World'
Full filter invocation with sendcmd would look like this:
sendcmd=c='56.0 drawtext reinit fontsize=56\:fontcolor=green\:text=Hello\\ World'
If the entire argument can't be parsed or applied as valid values then the filter will continue with its existing parameters.
Examples
drawtext="fontfile=/usr/share/fonts/truetype/freefont/FreeSerif.ttf: text='Test Text'"
drawtext="fontfile=/usr/share/fonts/truetype/freefont/FreeSerif.ttf: text='Test Text':\ x=100: y=50: fontsize=24: fontcolor=yellow@0.2: box=1: boxcolor=red@0.2"
Note that the double quotes are not necessary if spaces are not used within the parameter list.
drawtext="fontsize=30:fontfile=FreeSerif.ttf:text='hello world':x=(w-text_w)/2:y=(h-text_h)/2"
drawtext="fontsize=30:fontfile=FreeSerif.ttf:text='hello world':x=if(eq(mod(t\,30)\,0)\,rand(0\,(w-text_w))\,x):y=if(eq(mod(t\,30)\,0)\,rand(0\,(h-text_h))\,y)"
drawtext="fontsize=15:fontfile=FreeSerif.ttf:text=LONG_LINE:y=h-line_h:x=-50*t"
drawtext="fontsize=20:fontfile=FreeSerif.ttf:textfile=CREDITS:y=h-20*t"
drawtext="fontsize=60:fontfile=FreeSerif.ttf:fontcolor=green:text=g:x=(w-max_glyph_w)/2:y=h/2-ascent"
drawtext="fontfile=FreeSerif.ttf:fontcolor=white:x=100:y=x/dar:enable=lt(mod(t\,3)\,1):text='blink'"
drawtext='fontfile=Linux Libertine O-40\:style=Semibold:text=FFmpeg'
drawtext='fontfile=FreeSans.ttf:text=%{localtime\:%a %b %d %Y}'
#!/bin/sh DS=1.0 # display start DE=10.0 # display end FID=1.5 # fade in duration FOD=5 # fade out duration ffplay -f lavfi "color,drawtext=text=TEST:fontsize=50:fontfile=FreeSerif.ttf:fontcolor_expr=ff0000%{eif\\\\: clip(255*(1*between(t\\, $DS + $FID\\, $DE - $FOD) + ((t - $DS)/$FID)*between(t\\, $DS\\, $DS + $FID) + (-(t - $DE)/$FOD)*between(t\\, $DE - $FOD\\, $DE) )\\, 0\\, 255) \\\\: x\\\\: 2 }"
drawtext=fontfile=FreeSans.ttf:text=DOG:fontsize=24:x=10:y=20+24-max_glyph_a, drawtext=fontfile=FreeSans.ttf:text=cow:fontsize=24:x=80:y=20+24-max_glyph_a
drawtext="fontsize=20:fontcolor=white:fontfile=FreeSans.ttf:text='%{metadata\:lavf.image2dec.source_basename\:NA}':x=10:y=10"
For more information about libfreetype, check: <http://www.freetype.org/>.
For more information about fontconfig, check: <http://freedesktop.org/software/fontconfig/fontconfig-user.html>.
For more information about libfribidi, check: <http://fribidi.org/>.
Detect and draw edges. The filter uses the Canny Edge Detection algorithm.
The filter accepts the following options:
The high threshold selects the "strong" edge pixels, which are then connected through 8-connectivity with the "weak" edge pixels selected by the low threshold.
low and high threshold values must be chosen in the range [0,1], and low should be lesser or equal to high.
Default value for low is "20/255", and default value for high is "50/255".
Default value is wires.
Examples
edgedetect=low=0.1:high=0.4
edgedetect=mode=colormix:high=0
Apply a posterize effect using the ELBG (Enhanced LBG) algorithm.
For each input image, the filter will compute the optimal mapping from the input to the output given the codebook length, that is the number of distinct output colors.
This filter accepts the following options.
Measure graylevel entropy in histogram of color channels of video frames.
It accepts the following parameters:
diff mode measures entropy of histogram delta values, absolute differences between neighbour histogram values.
Set brightness, contrast, saturation and approximate gamma adjustment.
The filter accepts the following options:
It accepts the following values:
Default value is init.
The expressions accept the following parameters:
Commands
The filter supports the following commands:
The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Apply erosion effect to the video.
This filter replaces the pixel by the local(3x3) minimum.
It accepts the following options:
Flags to local 3x3 coordinates maps like this:
1 2 3 4 5 6 7 8
Commands
This filter supports the all above options as commands.
Extract color channel components from input video stream into separate grayscale video streams.
The filter accepts the following option:
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.
Examples
ffmpeg -i video.avi -filter_complex 'extractplanes=y+u+v[y][u][v]' -map '[y]' y.avi -map '[u]' u.avi -map '[v]' v.avi
Apply a fade-in/out effect to the input video.
It accepts the following parameters:
Examples
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
Denoise frames using 3D FFT (frequency domain filtering).
The filter accepts the following options:
Apply arbitrary expressions to samples in frequency domain
It accepts the following values:
Default value is init.
The filter accepts the following variables:
Examples
fftfilt=dc_Y=128:weight_Y='squish(1-(Y+X)/100)'
fftfilt=dc_Y=0:weight_Y='squish((Y+X)/100-1)'
fftfilt=dc_Y=0:weight_Y='1+squish(1-(Y+X)/100)'
fftfilt=dc_Y=0:weight_Y='exp(-4 * ((Y+X)/(W+H)))'
Extract a single field from an interlaced image using stride arithmetic to avoid wasting CPU time. The output frames are marked as non-interlaced.
The filter accepts the following options:
Create new frames by copying the top and bottom fields from surrounding frames supplied as numbers by the hint file.
There must be one line for each frame in a clip. Each line must contain two numbers separated by the comma, optionally followed by "-" or "+". Numbers supplied on each line of file can not be out of [N-1,N+1] where N is current frame number for "absolute" mode or out of [-1, 1] range for "relative" mode. First number tells from which frame to pick up top field and second number tells from which frame to pick up bottom field.
If optionally followed by "+" output frame will be marked as interlaced, else if followed by "-" output frame will be marked as progressive, else it will be marked same as input frame. If optionally followed by "t" output frame will use only top field, or in case of "b" it will use only bottom field. If line starts with "#" or ";" that line is skipped.
Example of first several lines of "hint" file for "relative" mode:
0,0 - # first frame 1,0 - # second frame, use third's frame top field and second's frame bottom field 1,0 - # third frame, use fourth's frame top field and third's frame bottom field 1,0 - 0,0 - 0,0 - 1,0 - 1,0 - 1,0 - 0,0 - 0,0 - 1,0 - 1,0 - 1,0 - 0,0 -
Field matching filter for inverse telecine. It is meant to reconstruct the progressive frames from a telecined stream. The filter does not drop duplicated frames, so to achieve a complete inverse telecine "fieldmatch" needs to be followed by a decimation filter such as decimate in the filtergraph.
The separation of the field matching and the decimation is notably motivated by the possibility of inserting a de-interlacing filter fallback between the two. If the source has mixed telecined and real interlaced content, "fieldmatch" will not be able to match fields for the interlaced parts. But these remaining combed frames will be marked as interlaced, and thus can be de-interlaced by a later filter such as yadif before decimation.
In addition to the various configuration options, "fieldmatch" can take an optional second stream, activated through the ppsrc option. If enabled, the frames reconstruction will be based on the fields and frames from this second stream. This allows the first input to be pre-processed in order to help the various algorithms of the filter, while keeping the output lossless (assuming the fields are matched properly). Typically, a field-aware denoiser, or brightness/contrast adjustments can help.
Note that this filter uses the same algorithms as TIVTC/TFM (AviSynth project) and VIVTC/VFM (VapourSynth project). The later is a light clone of TFM from which "fieldmatch" is based on. While the semantic and usage are very close, some behaviour and options names can differ.
The decimate filter currently only works for constant frame rate input. If your input has mixed telecined (30fps) and progressive content with a lower framerate like 24fps use the following filterchain to produce the necessary cfr stream: "dejudder,fps=30000/1001,fieldmatch,decimate".
The filter accepts the following options:
Note that it is sometimes recommended not to trust the parity announced by the stream.
Default value is auto.
More details about p/c/n/u/b are available in p/c/n/u/b meaning section.
Available values are:
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.
Default value is 0 (disabled).
Default value is auto.
Default value is 1.
Default value is 12.0.
Default is sc.
Default value is none.
Default value is 9.
Default value is 0.
Default value is 16.
Default value is 80.
p/c/n/u/b meaning
p/c/n
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:
u/b
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
Examples
Simple IVTC of a top field first telecined stream:
fieldmatch=order=tff:combmatch=none, decimate
Advanced IVTC, with fallback on yadif for still combed frames:
fieldmatch=order=tff:combmatch=full, yadif=deint=interlaced, decimate
Transform the field order of the input video.
It accepts the following parameters:
The default value is tff.
The transformation is done by shifting the picture content up or down by one line, and filling the remaining line with appropriate picture content. This method is consistent with most broadcast field order converters.
If the input video is not flagged as being interlaced, or it is already flagged as being of the required output field order, then this filter does not alter the incoming video.
It is very useful when converting to or from PAL DV material, which is bottom field first.
For example:
ffmpeg -i in.vob -vf "fieldorder=bff" out.dv
Buffer input images and send them when they are requested.
It is mainly useful when auto-inserted by the libavfilter framework.
It does not take parameters.
Fill borders of the input video, without changing video stream dimensions. Sometimes video can have garbage at the four edges and you may not want to crop video input to keep size multiple of some number.
This filter accepts the following options:
It accepts the following values:
Default is smear.
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Find a rectangular object
It accepts the following options:
Examples
ffmpeg -i file.ts -vf find_rect=newref.pgm,cover_rect=cover.jpg:mode=cover new.mkv
Flood area with values of same pixel components with another values.
It accepts the following options:
Convert the input video to one of the specified pixel formats. Libavfilter will try to pick one that is suitable as input to the next filter.
It accepts the following parameters:
Examples
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.
It accepts the following parameters:
Possible values are:
The default is "near".
Possible values are:
The default is "round".
Alternatively, the options can be specified as a flat string: fps[:start_time[:round]].
See also the setpts filter.
Examples
fps=fps=25
fps=fps=film:round=near
Pack two different video streams into a stereoscopic video, setting proper metadata on supported codecs. The two views should have the same size and framerate and processing will stop when the shorter video ends. Please note that you may conveniently adjust view properties with the scale and fps filters.
It accepts the following parameters:
Some examples:
# Convert left and right views into a frame-sequential video ffmpeg -i LEFT -i RIGHT -filter_complex framepack=frameseq OUTPUT # Convert views into a side-by-side video with the same output resolution as the input ffmpeg -i LEFT -i RIGHT -filter_complex [0:v]scale=w=iw/2[left],[1:v]scale=w=iw/2[right],[left][right]framepack=sbs OUTPUT
Change the frame rate by interpolating new video output frames from the source frames.
This filter is not designed to function correctly with interlaced media. If you wish to change the frame rate of interlaced media then you are required to deinterlace before this filter and re-interlace after this filter.
A description of the accepted options follows.
Available value for flags is:
Select one frame every N-th frame.
This filter accepts the following option:
Detect frozen video.
This filter logs a message and sets frame metadata when it detects that the input video has no significant change in content during a specified duration. Video freeze detection calculates the mean average absolute difference of all the components of video frames and compares it to a noise floor.
The printed times and duration are expressed in seconds. The "lavfi.freezedetect.freeze_start" metadata key is set on the first frame whose timestamp equals or exceeds the detection duration and it contains the timestamp of the first frame of the freeze. The "lavfi.freezedetect.freeze_duration" and "lavfi.freezedetect.freeze_end" metadata keys are set on the first frame after the freeze.
The filter accepts the following options:
Freeze video frames.
This filter freezes video frames using frame from 2nd input.
The filter accepts the following options:
Apply a frei0r effect to the input video.
To enable the compilation of this filter, you need to install the frei0r header and configure FFmpeg with "--enable-frei0r".
It accepts the following parameters:
A frei0r effect parameter can be a boolean (its value is either "y" or "n"), a double, a color (specified as R/G/B, where R, G, and B are floating point numbers between 0.0 and 1.0, inclusive) or a color description as specified in the "Color" section in the ffmpeg-utils manual, a position (specified as X/Y, where X and Y are floating point numbers) and/or a string.
The number and types of parameters depend on the loaded effect. If an effect parameter is not specified, the default value is set.
Examples
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>
Apply fast and simple postprocessing. It is a faster version of spp.
It splits (I)DCT into horizontal/vertical passes. Unlike the simple post- processing filter, one of them is performed once per block, not per pixel. This allows for much higher speed.
The filter accepts the following options:
Apply Gaussian blur filter.
The filter accepts the following options:
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Apply generic equation to each pixel.
The filter accepts the following options:
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:
Default is bilinear.
For functions, if x and y are outside the area, the value will be automatically clipped to the closer edge.
Please note that this filter can use multiple threads in which case each slice will have its own expression state. If you want to use only a single expression state because your expressions depend on previous state then you should limit the number of filter threads to 1.
Examples
geq=p(W-X\,Y)
geq=128 + 100*sin(2*(PI/100)*(cos(PI/3)*(X-50*T) + sin(PI/3)*Y)):128:128
nullsrc=s=256x256,geq=random(1)/hypot(X-cos(N*0.07)*W/2-W/2\,Y-sin(N*0.09)*H/2-H/2)^2*1000000*sin(N*0.02):128:128
format=gray,geq=lum_expr='(p(X,Y)+(256-p(X-4,Y-4)))/2'
geq=r='X/W*r(X,Y)':g='(1-X/W)*g(X,Y)':b='(H-Y)/H*b(X,Y)'
geq=lum=255*gauss((X/W-0.5)*3)*gauss((Y/H-0.5)*3)/gauss(0)/gauss(0),format=gray
Fix the banding artifacts that are sometimes introduced into nearly flat regions by truncation to 8-bit color depth. Interpolate the gradients that should go where the bands are, and dither them.
It is designed for playback only. Do not use it prior to lossy compression, because compression tends to lose the dither and bring back the bands.
It accepts the following parameters:
Alternatively, the options can be specified as a flat string: strength[:radius]
Examples
gradfun=3.5:8
gradfun=radius=8
Show various filtergraph stats.
With this filter one can debug complete filtergraph. Especially issues with links filling with queued frames.
The filter accepts the following options:
Available values for flags are:
A color constancy variation filter which estimates scene illumination via grey edge algorithm and corrects the scene colors accordingly.
See: <https://staff.science.uva.nl/th.gevers/pub/GeversTIP07.pdf>
The filter accepts the following options:
Examples
greyedge=difford=1:minknorm=5:sigma=2
greyedge=difford=1:minknorm=0:sigma=2
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:
"haldclut" also has the same interpolation options as lut3d (both filters share the same internals).
This filter also supports the framesync options.
More information about the Hald CLUT can be found on Eskil Steenberg's website (Hald CLUT author) at <http://www.quelsolaar.com/technology/clut.html>.
Workflow examples
Hald CLUT video stream
Generate an identity Hald CLUT stream altered with various effects:
ffmpeg -f lavfi -i B<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.
Hald CLUT with preview
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 B<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:
Compute and draw a color distribution histogram for the input video.
The computed histogram is a representation of the color component distribution in an image.
Standard histogram displays the color components distribution in an image. Displays color graph for each color component. Shows distribution of the Y, U, V, A or R, G, B components, depending on input format, in the current frame. Below each graph a color component scale meter is shown.
The filter accepts the following options:
Default is "stack".
Examples
ffplay -i input -vf histogram
This is a high precision/quality 3d denoise filter. It aims to reduce image noise, producing smooth images and making still images really still. It should enhance compressibility.
It accepts the following optional parameters:
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Download hardware frames to system memory.
The input must be in hardware frames, and the output a non-hardware format. Not all formats will be supported on the output - it may be necessary to insert an additional format filter immediately following in the graph to get the output in a supported format.
Map hardware frames to system memory or to another device.
This filter has several different modes of operation; which one is used depends on the input and output formats:
Map the input frames to system memory and pass them to the output. If the original hardware frame is later required (for example, after overlaying something else on part of it), the hwmap filter can be used again in the next mode to retrieve it.
If the input is actually a software-mapped hardware frame, then unmap it - that is, return the original hardware frame.
Otherwise, a device must be provided. Create new hardware surfaces on that device for the output, then map them back to the software format at the input and give those frames to the preceding filter. This will then act like the hwupload filter, but may be able to avoid an additional copy when the input is already in a compatible format.
A device must be supplied for the output, either directly or with the derive_device option. The input and output devices must be of different types and compatible - the exact meaning of this is system-dependent, but typically it means that they must refer to the same underlying hardware context (for example, refer to the same graphics card).
If the input frames were originally created on the output device, then unmap to retrieve the original frames.
Otherwise, map the frames to the output device - create new hardware frames on the output corresponding to the frames on the input.
The following additional parameters are accepted:
This may improve performance in some cases, as the original contents of the frame need not be loaded.
Indirect mappings to copies of frames are created in some cases where either direct mapping is not possible or it would have unexpected properties. Setting this flag ensures that the mapping is direct and will fail if that is not possible.
Defaults to read+write if not specified.
This option is dangerous - it may break the preceding filter in undefined ways if there are any additional constraints on that filter's output. Do not use it without fully understanding the implications of its use.
Upload system memory frames to hardware surfaces.
The device to upload to must be supplied when the filter is initialised. If using ffmpeg, select the appropriate device with the -filter_hw_device option or with the derive_device option. The input and output devices must be of different types and compatible - the exact meaning of this is system-dependent, but typically it means that they must refer to the same underlying hardware context (for example, refer to the same graphics card).
The following additional parameters are accepted:
Upload system memory frames to a CUDA device.
It accepts the following optional parameters:
Apply a high-quality magnification filter designed for pixel art. This filter was originally created by Maxim Stepin.
It accepts the following option:
Stack input videos horizontally.
All streams must be of same pixel format and of same height.
Note that this filter is faster than using overlay and pad filter to create same output.
The filter accepts the following option:
Modify the hue and/or the saturation of the input.
It accepts the following parameters:
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:
Examples
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))"
Commands
This filter supports the following commands:
Grow first stream into second stream by connecting components. This makes it possible to build more robust edge masks.
This filter accepts the following options:
The "hysteresis" filter also supports the framesync options.
Detect video interlacing type.
This filter tries to detect if the input frames are interlaced, progressive, top or bottom field first. It will also try to detect fields that are repeated between adjacent frames (a sign of telecine).
Single frame detection considers only immediately adjacent frames when classifying each frame. Multiple frame detection incorporates the classification history of previous frames.
The filter will log these metadata values:
The filter accepts the following options:
Deinterleave or interleave fields.
This filter allows one to process interlaced images fields without deinterlacing them. Deinterleaving splits the input frame into 2 fields (so called half pictures). Odd lines are moved to the top half of the output image, even lines to the bottom half. You can process (filter) them independently and then re-interleave them.
The filter accepts the following options:
Default value is "none".
Commands
This filter supports the all above options as commands.
Apply inflate effect to the video.
This filter replaces the pixel by the local(3x3) average by taking into account only values higher than the pixel.
It accepts the following options:
Commands
This filter supports the all above options as commands.
Simple interlacing filter from progressive contents. This interleaves upper (or lower) lines from odd frames with lower (or upper) lines from even frames, halving the frame rate and preserving image height.
Original Original New Frame Frame 'j' Frame 'j+1' (tff) ========== =========== ================== Line 0 --------------------> Frame 'j' Line 0 Line 1 Line 1 ----> Frame 'j+1' Line 1 Line 2 ---------------------> Frame 'j' Line 2 Line 3 Line 3 ----> Frame 'j+1' Line 3 ... ... ... New Frame + 1 will be generated by Frame 'j+2' and Frame 'j+3' and so on
It accepts the following optional parameters:
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.
Examples
kerndeint=thresh=10:map=0:order=0:sharp=0:twoway=0
kerndeint=sharp=1
kerndeint=map=1
Slowly update darker pixels.
This filter makes short flashes of light appear longer. This filter accepts the following options:
Correct radial lens distortion
This filter can be used to correct for radial distortion as can result from the use of wide angle lenses, and thereby re-rectify the image. To find the right parameters one can use tools available for example as part of opencv or simply trial-and-error. To use opencv use the calibration sample (under samples/cpp) from the opencv sources and extract the k1 and k2 coefficients from the resulting matrix.
Note that effectively the same filter is available in the open-source tools Krita and Digikam from the KDE project.
In contrast to the vignette filter, which can also be used to compensate lens errors, this filter corrects the distortion of the image, whereas vignette corrects the brightness distribution, so you may want to use both filters together in certain cases, though you will have to take care of ordering, i.e. whether vignetting should be applied before or after lens correction.
Options
The filter accepts the following options:
The formula that generates the correction is:
r_src = r_tgt * (1 + k1 * (r_tgt / r_0)^2 + k2 * (r_tgt / r_0)^4)
where r_0 is halve of the image diagonal and r_src and r_tgt are the distances from the focal point in the source and target images, respectively.
Apply lens correction via the lensfun library (<http://lensfun.sourceforge.net/>).
The "lensfun" filter requires the camera make, camera model, and lens model to apply the lens correction. The filter will load the lensfun database and query it to find the corresponding camera and lens entries in the database. As long as these entries can be found with the given options, the filter can perform corrections on frames. Note that incomplete strings will result in the filter choosing the best match with the given options, and the filter will output the chosen camera and lens models (logged with level "info"). You must provide the make, camera model, and lens model as they are required.
The filter accepts the following options:
Examples
ffmpeg -i input.mov -vf lensfun=make=Canon:model="Canon EOS 100D":lens_model="Canon EF-S 18-55mm f/3.5-5.6 IS STM":focal_length=18:aperture=8 -c:v h264 -b:v 8000k output.mov
ffmpeg -i input.mov -vf lensfun=make=Canon:model="Canon EOS 100D":lens_model="Canon EF-S 18-55mm f/3.5-5.6 IS STM":focal_length=18:aperture=8:enable='lte(t\,5)' -c:v h264 -b:v 8000k output.mov
Obtain the VMAF (Video Multi-Method Assessment Fusion) score between two input videos.
The obtained VMAF score is printed through the logging system.
It requires Netflix's vmaf library (libvmaf) as a pre-requisite. After installing the library it can be enabled using: "./configure --enable-libvmaf --enable-version3". If no model path is specified it uses the default model: "vmaf_v0.6.1.pkl".
The filter has following options:
This filter also supports the framesync options.
Examples
ffmpeg -i main.mpg -i ref.mpg -lavfi libvmaf -f null -
ffmpeg -i main.mpg -i ref.mpg -lavfi libvmaf="psnr=1:log_fmt=json" -f null -
ffmpeg -i main.mpg -i ref.mkv -lavfi "[0:v]settb=AVTB,setpts=PTS-STARTPTS[main];[1:v]settb=AVTB,setpts=PTS-STARTPTS[ref];[main][ref]libvmaf=psnr=1:log_fmt=json" -f null -
Limits the pixel components values to the specified range [min, max].
The filter accepts the following options:
Loop video frames.
The filter accepts the following options:
Examples
loop=loop=-1:size=1:start=0
loop=loop=10:size=1:start=0
loop=loop=5:size=10:start=0
Apply a 1D LUT to an input video.
The filter accepts the following options:
Currently supported formats:
Available values are:
Apply a 3D LUT to an input video.
The filter accepts the following options:
Currently supported formats:
Available values are:
Turn certain luma values into transparency.
The filter accepts the following options:
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Compute a look-up table for binding each pixel component input value to an output value, and apply it to the input video.
lutyuv applies a lookup table to a YUV input video, lutrgb to an RGB input video.
These filters accept the following parameters:
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:
All expressions default to "val".
Examples
lutrgb="r=maxval+minval-val:g=maxval+minval-val:b=maxval+minval-val" lutyuv="y=maxval+minval-val:u=maxval+minval-val:v=maxval+minval-val"
The above is the same as:
lutrgb="r=negval:g=negval:b=negval" lutyuv="y=negval:u=negval:v=negval"
lutyuv=y=negval
lutyuv="u=128:v=128"
lutyuv="y=2*val"
lutrgb="g=0:b=0"
format=rgba,lutrgb=a="maxval-minval/2"
lutyuv=y=gammaval(0.5)
lutyuv=y='bitand(val, 128+64+32)'
lutyuv=u='(val-maxval/2)*2+maxval/2':v='(val-maxval/2)*2+maxval/2'
The "lut2" filter takes two input streams and outputs one stream.
The "tlut2" (time lut2) filter takes two consecutive frames from one single stream.
This filter accepts the following parameters:
The "lut2" filter also supports the framesync options.
Each of them specifies the expression to use for computing the lookup table for the corresponding pixel component values.
The exact component associated to each of the c* options depends on the format in inputs.
The expressions can contain the following constants:
All expressions default to "x".
Examples
lut2='ifnot(x-y,0,pow(2,bdx)-1):ifnot(x-y,0,pow(2,bdx)-1):ifnot(x-y,0,pow(2,bdx)-1)'
lut2='ifnot(x-y,0,pow(2,bdx)-1):ifnot(x-y,pow(2,bdx-1),pow(2,bdx)-1):ifnot(x-y,pow(2,bdx-1),pow(2,bdx)-1)'
lut2='if(lt(x,y),0,if(gt(x,y),pow(2,bdx)-1,pow(2,bdx-1))):if(lt(x,y),0,if(gt(x,y),pow(2,bdx)-1,pow(2,bdx-1))):if(lt(x,y),0,if(gt(x,y),pow(2,bdx)-1,pow(2,bdx-1)))'
Clamp the first input stream with the second input and third input stream.
Returns the value of first stream to be between second input stream - "undershoot" and third input stream + "overshoot".
This filter accepts the following options:
Merge the second and third input stream into output stream using absolute differences between second input stream and first input stream and absolute difference between third input stream and first input stream. The picked value will be from second input stream if second absolute difference is greater than first one or from third input stream otherwise.
This filter accepts the following options:
Merge the first input stream with the second input stream using per pixel weights in the third input stream.
A value of 0 in the third stream pixel component means that pixel component from first stream is returned unchanged, while maximum value (eg. 255 for 8-bit videos) means that pixel component from second stream is returned unchanged. Intermediate values define the amount of merging between both input stream's pixel components.
This filter accepts the following options:
Merge the second and third input stream into output stream using absolute differences between second input stream and first input stream and absolute difference between third input stream and first input stream. The picked value will be from second input stream if second absolute difference is less than first one or from third input stream otherwise.
This filter accepts the following options:
Pick pixels comparing absolute difference of two video streams with fixed threshold.
If absolute difference between pixel component of first and second video stream is equal or lower than user supplied threshold than pixel component from first video stream is picked, otherwise pixel component from second video stream is picked.
This filter accepts the following options:
Create mask from input video.
For example it is useful to create motion masks after "tblend" filter.
This filter accepts the following options:
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:
It accepts one of the following values:
Default value is fast.
Default value is bff.
Higher values should result in a smoother motion vector field but less optimal individual vectors. Default value is 1.
Pick median pixel from certain rectangle defined by radius.
This filter accepts the following options:
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Merge color channel components from several video streams.
The filter accepts up to 4 input streams, and merge selected input planes to the output video.
This filter accepts the following options:
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.
Examples
[a0][a1][a2]mergeplanes=0x001020:yuv444p
[a0][a1]mergeplanes=0x00010210:yuva444p
format=yuva444p,mergeplanes=0x03010200:yuva444p
format=yuv420p,mergeplanes=0x000201:yuv420p
format=rgb24,mergeplanes=0x000102:yuv444p
Estimate and export motion vectors using block matching algorithms. Motion vectors are stored in frame side data to be used by other filters.
This filter accepts the following options:
Default value is esa.
Apply Midway Image Equalization effect using two video streams.
Midway Image Equalization adjusts a pair of images to have the same histogram, while maintaining their dynamics as much as possible. It's useful for e.g. matching exposures from a pair of stereo cameras.
This filter has two inputs and one output, which must be of same pixel format, but may be of different sizes. The output of filter is first input adjusted with midway histogram of both inputs.
This filter accepts the following option:
Convert the video to specified frame rate using motion interpolation.
This filter accepts the following options:
Default mode is obmc.
Default mode is bilat.
Default algorithm is epzs.
Default method is fdiff.
Mix several video input streams into one video stream.
A description of the accepted options follows.
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.
Default value is 0.
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 (invert) the input video.
It accepts the following option:
Denoise frames using Non-Local Means algorithm.
Each pixel is adjusted by looking for other pixels with similar contexts. This context similarity is defined by comparing their surrounding patches of size pxp. Patches are searched in an area of rxr around the pixel.
Note that the research area defines centers for patches, which means some patches will be made of pixels outside that research area.
The filter accepts the following options.
The default value is 0 and means automatic.
The default value is 0 and means automatic.
Deinterlace video using neural network edge directed interpolation.
This filter accepts the following options:
Can be one of the following:
Can be one of the following:
Can be one of the following:
Default is "new".
Force libavfilter not to use any of the specified pixel formats for the input to the next filter.
It accepts the following parameters:
Examples
noformat=pix_fmts=yuv420p,vflip
noformat=yuv420p|yuv444p|yuv410p
Add noise on video input frame.
The filter accepts the following options:
Examples
Add temporal and uniform noise to input video:
noise=alls=20:allf=t+u
Normalize RGB video (aka histogram stretching, contrast stretching). See: https://en.wikipedia.org/wiki/Normalization_(image_processing)
For each channel of each frame, the filter computes the input range and maps it linearly to the user-specified output range. The output range defaults to the full dynamic range from pure black to pure white.
Temporal smoothing can be used on the input range to reduce flickering (rapid changes in brightness) caused when small dark or bright objects enter or leave the scene. This is similar to the auto-exposure (automatic gain control) on a video camera, and, like a video camera, it may cause a period of over- or under-exposure of the video.
The R,G,B channels can be normalized independently, which may cause some color shifting, or linked together as a single channel, which prevents color shifting. Linked normalization preserves hue. Independent normalization does not, so it can be used to remove some color casts. Independent and linked normalization can be combined in any ratio.
The normalize filter accepts the following options:
Commands
This filter supports same commands as options, excluding smoothing option. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Examples
Stretch video contrast to use the full dynamic range, with no temporal smoothing; may flicker depending on the source content:
normalize=blackpt=black:whitept=white:smoothing=0
As above, but with 50 frames of temporal smoothing; flicker should be reduced, depending on the source content:
normalize=blackpt=black:whitept=white:smoothing=50
As above, but with hue-preserving linked channel normalization:
normalize=blackpt=black:whitept=white:smoothing=50:independence=0
As above, but with half strength:
normalize=blackpt=black:whitept=white:smoothing=50:independence=0:strength=0.5
Map the darkest input color to red, the brightest input color to cyan:
normalize=blackpt=red:whitept=cyan
Pass the video source unchanged to the output.
Optical Character Recognition
This filter uses Tesseract for optical character recognition. To enable compilation of this filter, you need to configure FFmpeg with "--enable-libtesseract".
It accepts the following options:
The filter exports recognized text as the frame metadata "lavfi.ocr.text". The filter exports confidence of recognized words as the frame metadata "lavfi.ocr.confidence".
Apply a video transform using libopencv.
To enable this filter, install the libopencv library and headers and configure FFmpeg with "--enable-libopencv".
It accepts the following parameters:
Refer to the official libopencv documentation for more precise information: <http://docs.opencv.org/master/modules/imgproc/doc/filtering.html>
Several libopencv filters are supported; see the following subsections.
dilate
Dilate an image by using a specific structuring element. It corresponds to the libopencv function "cvDilate".
It accepts the parameters: struct_el|nb_iterations.
struct_el represents a structuring element, and has the syntax: colsxrows+anchor_xxanchor_y/shape
cols and rows represent the number of columns and rows of the structuring element, anchor_x and anchor_y the anchor point, and shape the shape for the structuring element. shape must be "rect", "cross", "ellipse", or "custom".
If the value for shape is "custom", it must be followed by a string of the form "=filename". The file with name filename is assumed to represent a binary image, with each printable character corresponding to a bright pixel. When a custom shape is used, cols and rows are ignored, the number or columns and rows of the read file are assumed instead.
The default value for struct_el is "3x3+0x0/rect".
nb_iterations specifies the number of times the transform is applied to the image, and defaults to 1.
Some examples:
# Use the default values ocv=dilate # Dilate using a structuring element with a 5x5 cross, iterating two times ocv=filter_name=dilate:filter_params=5x5+2x2/cross|2 # Read the shape from the file diamond.shape, iterating two times. # The file diamond.shape may contain a pattern of characters like this # * # *** # ***** # *** # * # The specified columns and rows are ignored # but the anchor point coordinates are not ocv=dilate:0x0+2x2/custom=diamond.shape|2
erode
Erode an image by using a specific structuring element. It corresponds to the libopencv function "cvErode".
It accepts the parameters: struct_el:nb_iterations, with the same syntax and semantics as the dilate filter.
smooth
Smooth the input video.
The filter takes the following parameters: type|param1|param2|param3|param4.
type is the type of smooth filter to apply, and must be one of the following values: "blur", "blur_no_scale", "median", "gaussian", or "bilateral". The default value is "gaussian".
The meaning of param1, param2, param3, and param4 depends on the smooth type. param1 and param2 accept integer positive values or 0. param3 and param4 accept floating point values.
The default value for param1 is 3. The default value for the other parameters is 0.
These parameters correspond to the parameters assigned to the libopencv function "cvSmooth".
2D Video Oscilloscope.
Useful to measure spatial impulse, step responses, chroma delays, etc.
It accepts the following parameters:
Commands
This filter supports same commands as options. The command accepts the same syntax of the corresponding option.
If the specified expression is not valid, it is kept at its current value.
Examples
oscilloscope=x=0.5:y=0:s=1
oscilloscope=x=0.5:y=1:s=1
oscilloscope=x=0.5:y=5/1080:s=1
oscilloscope=x=1:y=0.5:s=1:t=1
Overlay one video on top of another.
It takes two inputs and has one output. The first input is the "main" video on which the second input is overlaid.
It accepts the following parameters:
A description of the accepted options follows.
It accepts the following values:
Default value is frame.
It accepts the following values:
Default value is yuv420.
The x, and y expressions can contain the following parameters.
This filter also supports the framesync options.
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 the example for the movie filter does.
You can chain together more overlays but you should test the efficiency of such approach.
Commands
This filter supports the following commands:
If the specified expression is not valid, it is kept at its current value.
Examples
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 -i input -i logo -filter_complex 'overlay=10:main_h-overlay_h-10' output
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
color=color=red@.3:size=WxH [over]; [in][over] overlay [out]
ffplay input.avi -vf 'split[a][b]; [a]pad=iw*2:ih[src]; [b]deshake[filt]; [src][filt]overlay=w'
The above command is the same as:
ffplay input.avi -vf 'split[b], pad=iw*2[src], [b]deshake, [src]overlay=w'
overlay=x='if(gte(t,2), -w+(t-2)*20, NAN)':y=0
ffmpeg -i left.avi -i right.avi -filter_complex " nullsrc=size=200x100 [background]; [0:v] setpts=PTS-STARTPTS, scale=100x100 [left]; [1:v] setpts=PTS-STARTPTS, scale=100x100 [right]; [background][left] overlay=shortest=1 [background+left]; [background+left][right] overlay=shortest=1:x=100 [left+right] "
ffmpeg -i test.avi -codec:v:0 wmv2 -ar 11025 -b:v 9000k -vf '[in]split[split_main][split_delogo];[split_delogo]trim=start=360:end=371,delogo=0:0:640:480[delogoed];[split_main][delogoed]overlay=eof_action=pass[out]' masked.avi
nullsrc=s=200x200 [bg]; testsrc=s=100x100, split=4 [in0][in1][in2][in3]; [in0] lutrgb=r=0, [bg] overlay=0:0 [mid0]; [in1] lutrgb=g=0, [mid0] overlay=100:0 [mid1]; [in2] lutrgb=b=0, [mid1] overlay=0:100 [mid2]; [in3] null, [mid2] overlay=100:100 [out0]
Overlay one video on top of another.
This is the CUDA cariant of the overlay filter. It only accepts CUDA frames. The underlying input pixel formats have to match.
It takes two inputs and has one output. The first input is the "main" video on which the second input is overlaid.
It accepts the following parameters:
This filter also supports the framesync options.
Apply Overcomplete Wavelet denoiser.
The filter accepts the following options:
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.
Must be a double value in the range 0-1000, default is 1.0.
Must be a double value in the range 0-1000, default is 1.0.
Add paddings to the input image, and place the original input at the provided x, y coordinates.
It accepts the following parameters:
The width expression can reference the value set by the height expression, and vice versa.
The default value of width and height is 0.
The x expression can reference the value set by the y expression, and vice versa.
The default value of x and y is 0.
If x or y evaluate to a negative number, they'll be changed so the input image is centered on the padded area.
The default value of color is "black".
It accepts the following values:
Default value is init.
The value for the width, height, x, and y options are expressions containing the following constants:
Examples
pad=640:480:0:40:violet
The example above is equivalent to the following command:
pad=width=640:height=480:x=0:y=40:color=violet
pad="3/2*iw:3/2*ih:(ow-iw)/2:(oh-ih)/2"
pad="max(iw\,ih):ow:(ow-iw)/2:(oh-ih)/2"
pad="ih*16/9:ih:(ow-iw)/2:(oh-ih)/2"
(ih * X / ih) * sar = output_dar X = output_dar / sar
Thus the previous example needs to be modified to:
pad="ih*16/9/sar:ih:(ow-iw)/2:(oh-ih)/2"
pad="2*iw:2*ih:ow-iw:oh-ih"
Generate one palette for a whole video stream.
It accepts the following options:
It accepts the following values:
Default value is full.
The filter also exports the frame metadata "lavfi.color_quant_ratio" ("nb_color_in / nb_color_out") which you can use to evaluate the degree of color quantization of the palette. This information is also visible at info logging level.
Examples
ffmpeg -i input.mkv -vf palettegen palette.png
Use a palette to downsample an input video stream.
The filter takes two inputs: one video stream and a palette. The palette must be a 256 pixels image.
It accepts the following options:
Default is sierra2_4a.
The option must be an integer value in the range [0,5]. Default is 2.
Default is none.
The option must be an integer value in the range [0,255]. Default is 128.
Examples
ffmpeg -i input.mkv -i palette.png -lavfi paletteuse output.gif
Correct perspective of video not recorded perpendicular to the screen.
A description of the accepted parameters follows.
The expressions can use the following variables:
It accepts the following values:
Default value is linear.
It accepts the following values:
Default value is source.
It accepts the following values:
Default value is init.
Delay interlaced video by one field time so that the field order changes.
The intended use is to fix PAL movies that have been captured with the opposite field order to the film-to-video transfer.
A description of the accepted parameters follows.
It accepts the following values:
Reduce various flashes in video, so to help users with epilepsy.
It accepts the following options:
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.
Display sample values of color channels. Mainly useful for checking color and levels. Minimum supported resolution is 640x480.
The filters accept the following options:
Enable the specified chain of postprocessing subfilters using libpostproc. This library should be automatically selected with a GPL build ("--enable-gpl"). Subfilters must be separated by '/' and can be disabled by prepending a '-'. Each subfilter and some options have a short and a long name that can be used interchangeably, i.e. dr/dering are the same.
The filters accept the following options:
All subfilters share common options to determine their scope:
These options can be appended after the subfilter name, separated by a '|'.
Available subfilters are:
The horizontal and vertical deblocking filters share the difference and flatness values so you cannot set different horizontal and vertical thresholds.
Examples
pp=hb/vb/dr/al
pp=de/-al
pp=default/tmpnoise|1|2|3
pp=hb|y/vb|a
Apply Postprocessing filter 7. It is variant of the spp filter, similar to spp = 6 with 7 point DCT, where only the center sample is used after IDCT.
The filter accepts the following options:
Apply alpha premultiply effect to input video stream using first plane of second stream as alpha.
Both streams must have same dimensions and same pixel format.
The filter accepts the following option:
Apply prewitt operator to input video stream.
The filter accepts the following option:
Alter frame colors in video with pseudocolors.
This filter accepts the following options:
Each of them specifies the expression to use for computing the lookup table for the corresponding pixel component values.
The expressions can contain the following constants and functions:
All expressions default to "val".
Examples
pseudocolor="'if(between(val,ymax,amax),lerp(ymin,ymax,(val-ymax)/(amax-ymax)),-1):if(between(val,ymax,amax),lerp(umax,umin,(val-ymax)/(amax-ymax)),-1):if(between(val,ymax,amax),lerp(vmin,vmax,(val-ymax)/(amax-ymax)),-1):-1'"
Obtain the average, maximum and minimum PSNR (Peak Signal to Noise Ratio) between two input videos.
This filter takes in input two input videos, the first input is considered the "main" source and is passed unchanged to the output. The second input is used as a "reference" video for computing the PSNR.
Both video inputs must have the same resolution and pixel format for this filter to work correctly. Also it assumes that both inputs have the same number of frames, which are compared one by one.
The obtained average PSNR is printed through the logging system.
The filter stores the accumulated MSE (mean squared error) of each frame, and at the end of the processing it is averaged across all frames equally, and the following formula is applied to obtain the PSNR:
PSNR = 10*log10(MAX^2/MSE)
Where MAX is the average of the maximum values of each component of the image.
The description of the accepted parameters follows.
This filter also supports the framesync options.
The file printed if stats_file is selected, contains a sequence of key/value pairs of the form key:value for each compared couple of frames.
If a stats_version greater than 1 is specified, a header line precedes the list of per-frame-pair stats, with key value pairs following the frame format with the following parameters:
A description of each shown per-frame-pair parameter follows:
Examples
movie=ref_movie.mpg, setpts=PTS-STARTPTS [main]; [main][ref] psnr="stats_file=stats.log" [out]
On this example the input file being processed is compared with the reference file ref_movie.mpg. The PSNR of each individual frame is stored in stats.log.
ffmpeg -i main.mpg -i ref.mkv -lavfi "[0:v]settb=AVTB,setpts=PTS-STARTPTS[main];[1:v]settb=AVTB,setpts=PTS-STARTPTS[ref];[main][ref]psnr" -f null -
Pulldown reversal (inverse telecine) filter, capable of handling mixed hard-telecine, 24000/1001 fps progressive, and 30000/1001 fps progressive content.
The pullup filter is designed to take advantage of future context in making its decisions. This filter is stateless in the sense that it does not lock onto a pattern to follow, but it instead looks forward to the following fields in order to identify matches and rebuild progressive frames.
To produce content with an even framerate, insert the fps filter after pullup, use "fps=24000/1001" if the input frame rate is 29.97fps, "fps=24" for 30fps and the (rare) telecined 25fps input.
The filter accepts the following options:
This option may be set to use chroma plane instead of the default luma plane for doing filter's computations. This may improve accuracy on very clean source material, but more likely will decrease accuracy, especially if there is chroma noise (rainbow effect) or any grayscale video. The main purpose of setting mp to a chroma plane is to reduce CPU load and make pullup usable in realtime on slow machines.
For best results (without duplicated frames in the output file) it is necessary to change the output frame rate. For example, to inverse telecine NTSC input:
ffmpeg -i input -vf pullup -r 24000/1001 ...
Change video quantization parameters (QP).
The filter accepts the following option:
The expression is evaluated through the eval API and can contain, among others, the following constants:
Examples
qp=2+2*sin(PI*qp)
Flush video frames from internal cache of frames into a random order. No frame is discarded. Inspired by frei0r nervous filter.
Read closed captioning (EIA-608) information from the top lines of a video frame.
This filter adds frame metadata for "lavfi.readeia608.X.cc" and "lavfi.readeia608.X.line", where "X" is the number of the identified line with EIA-608 data (starting from 0). A description of each metadata value follows:
This filter accepts the following options:
Examples
ffprobe -f lavfi -i movie=captioned_video.mov,readeia608 -show_entries frame=pkt_pts_time:frame_tags=lavfi.readeia608.0.cc,lavfi.readeia608.1.cc -of csv
Read vertical interval timecode (VITC) information from the top lines of a video frame.
The filter adds frame metadata key "lavfi.readvitc.tc_str" with the timecode value, if a valid timecode has been detected. Further metadata key "lavfi.readvitc.found" is set to 0/1 depending on whether timecode data has been found or not.
This filter accepts the following options:
Examples
ffmpeg -i input.avi -filter:v 'readvitc,drawtext=fontfile=FreeMono.ttf:text=%{metadata\\:lavfi.readvitc.tc_str\\:--\\\\\\:--\\\\\\:--\\\\\\:--}:x=(w-tw)/2:y=400-ascent'
Remap pixels using 2nd: Xmap and 3rd: Ymap input video stream.
Destination pixel at position (X, Y) will be picked from source (x, y) position where x = Xmap(X, Y) and y = Ymap(X, Y). If mapping values are out of range, zero value for pixel will be used for destination pixel.
Xmap and Ymap input video streams must be of same dimensions. Output video stream will have Xmap/Ymap video stream dimensions. Xmap and Ymap input video streams are 16bit depth, single channel.
The removegrain filter is a spatial denoiser for progressive video.
Range of mode is from 0 to 24. Description of each mode follows:
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:
Pixels in the provided bitmap image with a value of zero are not considered part of the logo, non-zero pixels are considered part of the logo. If you use white (255) for the logo and black (0) for the rest, you will be safe. For making the filter bitmap, it is recommended to take a screen capture of a black frame with the logo visible, and then using a threshold filter followed by the erode filter once or twice.
If needed, little splotches can be fixed manually. Remember that if logo pixels are not covered, the filter quality will be much reduced. Marking too many pixels as part of the logo does not hurt as much, but it will increase the amount of blurring needed to cover over the image and will destroy more information than necessary, and extra pixels will slow things down on a large logo.
This filter uses the repeat_field flag from the Video ES headers and hard repeats fields based on its value.
Reverse a video clip.
Warning: This filter requires memory to buffer the entire clip, so trimming is suggested.
Examples
trim=end=5,reverse
Shift R/G/B/A pixels horizontally and/or vertically.
The filter accepts the following options:
Commands
This filter supports the all above options as commands.
Apply roberts cross operator to input video stream.
The filter accepts the following option:
Rotate video by an arbitrary angle expressed in radians.
The filter accepts the following options:
A description of the optional parameters follows.
This expression is evaluated for each frame.
Default value is "black".
The expressions for the angle and the output size can contain the following constants and functions:
These are only available when computing the out_w and out_h expressions.
Examples
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
Commands
The filter supports the following commands:
If the specified expression is not valid, it is kept at its current value.
Apply Shape Adaptive Blur.
The filter accepts the following options:
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.
Options
The filter accepts the following options, or any of the options supported by the libswscale scaler.
See the ffmpeg-scaler manual for the complete list of scaler options.
If the width or w value is 0, the input width is used for the output. If the height or h value is 0, the input height is used for the output.
If one and only one of the values is -n with n >= 1, the scale filter will use a value that maintains the aspect ratio of the input image, calculated from the other specified dimension. After that it will, however, make sure that the calculated dimension is divisible by n and adjust the value if necessary.
If both values are -n with n >= 1, the behavior will be identical to both values being set to 0 as previously detailed.
See below for the list of accepted constants for use in the dimension expression.
Default value is init.
Default value is 0.
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:
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:
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.
This option respects the value set for force_original_aspect_ratio, increasing or decreasing the resolution accordingly. The video's aspect ratio may be slightly modified.
This option can be handy if you need to have a video fit within or exceed a defined resolution using force_original_aspect_ratio but also have encoder restrictions on width or height divisibility.
The values of the w and h options are expressions containing the following constants:
Examples
scale=w=200:h=100
This is equivalent to:
scale=200:100
or:
scale=200x100
scale=qcif
which can also be written as:
scale=size=qcif
scale=w=2*iw:h=2*ih
scale=2*in_w:2*in_h
scale=2*iw:2*ih:interl=1
scale=w=iw/2:h=ih/2
scale=3/2*iw:ow
scale=iw:1/PHI*iw scale=ih*PHI:ih
scale=w=3/2*oh:h=3/5*ih
scale="trunc(3/2*iw/hsub)*hsub:trunc(3/2*ih/vsub)*vsub"
scale=w='min(500\, iw*3/2):h=-1'
scale='trunc(ih*dar):ih',setsar=1/1
scale='trunc(ih*dar/2)*2:trunc(ih/2)*2',setsar=1/1
Commands
This filter supports the following commands:
Use the NVIDIA Performance Primitives (libnpp) to perform scaling and/or pixel format conversion on CUDA video frames. Setting the output width and height works in the same way as for the scale filter.
The following additional options are accepted:
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.
This option respects the value set for force_original_aspect_ratio, increasing or decreasing the resolution accordingly. The video's aspect ratio may be slightly modified.
This option can be handy if you need to have a video fit within or exceed a defined resolution using force_original_aspect_ratio but also have encoder restrictions on width or height divisibility.
Scale (resize) the input video, based on a reference video.
See the scale filter for available options, scale2ref supports the same but uses the reference video instead of the main input as basis. scale2ref also supports the following additional constants for the w and h options:
Examples
'scale2ref[b][a];[a][b]overlay'
[logo-in][video-in]scale2ref=w=oh*mdar:h=ih/10[logo-out][video-out]
Commands
This filter supports the following commands:
Scroll input video horizontally and/or vertically by constant speed.
The filter accepts the following options:
Commands
This filter supports the following commands:
Detect video scene change.
This filter sets frame metadata with mafd between frame, the scene score, and forward the frame to the next filter, so they can use these metadata to detect scene change or others.
In addition, this filter logs a message and sets frame metadata when it detects a scene change by threshold.
"lavfi.scd.mafd" metadata keys are set with mafd for every frame.
"lavfi.scd.score" metadata keys are set with scene change score for every frame to detect scene change.
"lavfi.scd.time" metadata keys are set with current filtered frame time which detect scene change with threshold.
The filter accepts the following options:
Default value is 10..
Adjust cyan, magenta, yellow and black (CMYK) to certain ranges of colors (such as "reds", "yellows", "greens", "cyans", ...). The adjustment range is defined by the "purity" of the color (that is, how saturated it already is).
This filter is similar to the Adobe Photoshop Selective Color tool.
The filter accepts the following options:
Available values are:
Default is "absolute".
All the adjustment settings (reds, yellows, ...) accept up to 4 space separated floating point adjustment values in the [-1,1] range, respectively to adjust the amount of cyan, magenta, yellow and black for the pixels of its range.
Examples
selectivecolor=greens=.5 0 -.33 0:blues=0 .27
selectivecolor=psfile=MySelectiveColorPresets/Misty.asv
The "separatefields" takes a frame-based video input and splits each frame into its components fields, producing a new half height clip with twice the frame rate and twice the frame count.
This filter use field-dominance information in frame to decide which of each pair of fields to place first in the output. If it gets it wrong use setfield filter before "separatefields" filter.
The "setdar" filter sets the Display Aspect Ratio for the filter output video.
This is done by changing the specified Sample (aka Pixel) Aspect Ratio, according to the following equation:
<DAR> = <HORIZONTAL_RESOLUTION> / <VERTICAL_RESOLUTION> * <SAR>
Keep in mind that the "setdar" filter does not modify the pixel dimensions of the video frame. Also, the display aspect ratio set by this filter may be changed by later filters in the filterchain, e.g. in case of scaling or if another "setdar" or a "setsar" filter is applied.
The "setsar" filter sets the Sample (aka Pixel) Aspect Ratio for the filter output video.
Note that as a consequence of the application of this filter, the output display aspect ratio will change according to the equation above.
Keep in mind that the sample aspect ratio set by the "setsar" filter may be changed by later filters in the filterchain, e.g. if another "setsar" or a "setdar" filter is applied.
It accepts the following parameters:
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.
The parameter sar is an expression containing the following constants:
Examples
setdar=dar=1.77777 setdar=dar=16/9
setsar=sar=10/11
setdar=ratio=16/9:max=1000
Force field for the output video frame.
The "setfield" filter marks the interlace type field for the output frames. It does not change the input frame, but only sets the corresponding property, which affects how the frame is treated by following filters (e.g. "fieldorder" or "yadif").
The filter accepts the following options:
Force frame parameter for the output video frame.
The "setparams" filter marks interlace and color range for the output frames. It does not change the input frame, but only sets the corresponding property, which affects how the frame is treated by filters/encoders.
Show a line containing various information for each input video frame. The input video is not modified.
This filter supports the following options:
The shown line contains a sequence of key/value pairs of the form key:value.
The following values are shown in the output:
Displays the 256 colors palette of each frame. This filter is only relevant for pal8 pixel format frames.
It accepts the following option:
Reorder and/or duplicate and/or drop video frames.
It accepts the following parameters:
The first frame has the index 0. The default is to keep the input unchanged.
Examples
ffmpeg -i INPUT -vf "shuffleframes=0 2 1" OUTPUT
ffmpeg -i INPUT -vf "shuffleframes=9 1 2 3 4 5 6 7 8 0" OUTPUT
Reorder and/or duplicate video planes.
It accepts the following parameters:
The first plane has the index 0. The default is to keep the input unchanged.
Examples
ffmpeg -i INPUT -vf shuffleplanes=0:2:1:3 OUTPUT
Evaluate various visual metrics that assist in determining issues associated with the digitization of analog video media.
By default the filter will log these metadata values:
The filter accepts the following options:
Both options accept the following values:
Examples
ffprobe -f lavfi movie=example.mov,signalstats="stat=tout+vrep+brng" -show_frames
ffprobe -f lavfi movie=example.mov,signalstats -show_entries frame_tags=lavfi.signalstats.YMAX,lavfi.signalstats.YMIN
ffplay example.mov -vf signalstats="out=brng:color=red"
ffplay example.mov -vf signalstats=stat=brng+vrep+tout,drawtext=fontfile=FreeSerif.ttf:textfile=signalstat_drawtext.txt
The contents of signalstat_drawtext.txt used in the command are:
time %{pts:hms} Y (%{metadata:lavfi.signalstats.YMIN}-%{metadata:lavfi.signalstats.YMAX}) U (%{metadata:lavfi.signalstats.UMIN}-%{metadata:lavfi.signalstats.UMAX}) V (%{metadata:lavfi.signalstats.VMIN}-%{metadata:lavfi.signalstats.VMAX}) saturation maximum: %{metadata:lavfi.signalstats.SATMAX}
Calculates the MPEG-7 Video Signature. The filter can handle more than one input. In this case the matching between the inputs can be calculated additionally. The filter always passes through the first input. The signature of each stream can be written into a file.
It accepts the following options:
Available values are:
Available values are:
Examples
ffmpeg -i input.mkv -vf signature=filename=signature.bin -map 0:v -f null -
ffmpeg -i input1.mkv -i input2.mkv -filter_complex "[0:v][1:v] signature=nb_inputs=2:detectmode=full:format=xml:filename=signature%d.xml" -map :v -f null -
Blur the input video without impacting the outlines.
It accepts the following options:
If a chroma option is not explicitly set, the corresponding luma value is set.
Apply sobel operator to input video stream.
The filter accepts the following option:
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:
Commands
This filter supports the following commands:
Scale the input by applying one of the super-resolution methods based on convolutional neural networks. Supported models:
Training scripts as well as scripts for model file (.pb) saving can be found at <https://github.com/XueweiMeng/sr/tree/sr_dnn_native>. Original repository is at <https://github.com/HighVoltageRocknRoll/sr.git>.
Native model files (.model) can be generated from TensorFlow model files (.pb) by using tools/python/convert.py
The filter accepts the following options:
Default value is native.
This feature can also be finished with dnn_processing filter.
Obtain the SSIM (Structural SImilarity Metric) between two input videos.
This filter takes in input two input videos, the first input is considered the "main" source and is passed unchanged to the output. The second input is used as a "reference" video for computing the SSIM.
Both video inputs must have the same resolution and pixel format for this filter to work correctly. Also it assumes that both inputs have the same number of frames, which are compared one by one.
The filter stores the calculated SSIM of each frame.
The description of the accepted parameters follows.
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:
This filter also supports the framesync options.
Examples
movie=ref_movie.mpg, setpts=PTS-STARTPTS [main]; [main][ref] ssim="stats_file=stats.log" [out]
On this example the input file being processed is compared with the reference file ref_movie.mpg. The SSIM of each individual frame is stored in stats.log.
ffmpeg -i main.mpg -i ref.mpg -lavfi "ssim;[0:v][1:v]psnr" -f null -
ffmpeg -i main.mpg -i ref.mkv -lavfi "[0:v]settb=AVTB,setpts=PTS-STARTPTS[main];[1:v]settb=AVTB,setpts=PTS-STARTPTS[ref];[main][ref]ssim" -f null -
Convert between different stereoscopic image formats.
The filters accept the following options:
Available values for input image formats are:
Default value is sbsl.
Default value is arcd.
Examples
stereo3d=sbsl:aybd
stereo3d=abl:sbsr
Select video or audio streams.
The filter accepts the following options:
Commands
The "streamselect" and "astreamselect" filter supports the following commands:
Examples
sendcmd='5.0 streamselect map 1',streamselect=inputs=2:map=0
asendcmd='5.0 astreamselect map 1',astreamselect=inputs=2:map=0
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:
If the first key is not specified, it is assumed that the first value specifies the filename.
For example, to render the file sub.srt on top of the input video, use the command:
subtitles=sub.srt
which is equivalent to:
subtitles=filename=sub.srt
To render the default subtitles stream from file video.mkv, use:
subtitles=video.mkv
To render the second subtitles stream from that file, use:
subtitles=video.mkv:si=1
To make the subtitles stream from sub.srt appear in 80% transparent blue "DejaVu Serif", use:
subtitles=sub.srt:force_style='FontName=DejaVu Serif,PrimaryColour=&HCCFF0000'
Scale the input by 2x and smooth using the Super2xSaI (Scale and Interpolate) pixel art scaling algorithm.
Useful for enlarging pixel art images without reducing sharpness.
Swap two rectangular objects in video.
This filter accepts the following options:
All expressions are evaluated once for each frame.
The all options are expressions containing the following constants:
Swap U & V plane.
Blend successive video frames.
See blend
Apply telecine process to the video.
This filter accepts the following options:
Some typical patterns: NTSC output (30i): 27.5p: 32222 24p: 23 (classic) 24p: 2332 (preferred) 20p: 33 18p: 334 16p: 3444 PAL output (25i): 27.5p: 12222 24p: 222222222223 ("Euro pulldown") 16.67p: 33 16p: 33333334
Compute and draw a color distribution histogram for the input video across time.
Unlike histogram video filter which only shows histogram of single input frame at certain time, this filter shows also past histograms of number of frames defined by "width" option.
The computed histogram is a representation of the color component distribution in an image.
The filter accepts the following options:
Default is "stack".
Apply threshold effect to video stream.
This filter needs four video streams to perform thresholding. First stream is stream we are filtering. Second stream is holding threshold values, third stream is holding min values, and last, fourth stream is holding max values.
The filter accepts the following option:
For example if first stream pixel's component value is less then threshold value of pixel component from 2nd threshold stream, third stream value will picked, otherwise fourth stream pixel component value will be picked.
Using color source filter one can perform various types of thresholding:
Examples
ffmpeg -i 320x240.avi -f lavfi -i color=gray -f lavfi -i color=black -f lavfi -i color=white -lavfi threshold output.avi
ffmpeg -i 320x240.avi -f lavfi -i color=gray -f lavfi -i color=white -f lavfi -i color=black -lavfi threshold output.avi
ffmpeg -i 320x240.avi -f lavfi -i color=gray -i 320x240.avi -f lavfi -i color=gray -lavfi threshold output.avi
ffmpeg -i 320x240.avi -f lavfi -i color=gray -f lavfi -i color=white -i 320x240.avi -lavfi threshold output.avi
ffmpeg -i 320x240.avi -f lavfi -i color=gray -i 320x240.avi -f lavfi -i color=white -lavfi threshold output.avi
Select the most representative frame in a given sequence of consecutive frames.
The filter accepts the following options:
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.
Examples
thumbnail=50
ffmpeg -i in.avi -vf thumbnail,scale=300:200 -frames:v 1 out.png
Tile several successive frames together.
The untile filter can do the reverse.
The filter accepts the following options:
Examples
ffmpeg -skip_frame nokey -i file.avi -vf 'scale=128:72,tile=8x8' -an -vsync 0 keyframes%03d.png
The -vsync 0 is necessary to prevent ffmpeg from duplicating each output frame to accommodate the originally detected frame rate.
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:
Available values are:
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 11111 33333 22222 44444 11111 33333 22222 44444 11111 33333 22222 44444 11111 33333 22222 44444
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 11111 33333 11111 33333 11111 33333 11111 33333
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 22222 44444 22222 44444 22222 44444 22222 44444
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 11111 ..... 33333 ..... ..... 22222 ..... 44444 11111 ..... 33333 ..... ..... 22222 ..... 44444 11111 ..... 33333 ..... ..... 22222 ..... 44444 11111 ..... 33333 ..... ..... 22222 ..... 44444
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111<- 22222 33333<- 44444 11111 22222<- 33333 44444<- 11111<- 22222 33333<- 44444 11111 22222<- 33333 44444<- Output: 11111 33333 22222 44444 11111 33333 22222 44444
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222<- 33333 44444<- 11111<- 22222 33333<- 44444 11111 22222<- 33333 44444<- 11111<- 22222 33333<- 44444 Output: 22222 44444 11111 33333 22222 44444 11111 33333
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 11111 22222 22222 33333 33333 44444 44444 11111 11111 22222 22222 33333 33333 44444 11111 22222 22222 33333 33333 44444 44444 11111 11111 22222 22222 33333 33333 44444
------> time Input: Frame 1 Frame 2 Frame 3 Frame 4 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 11111 22222 33333 44444 Output: 11111 33333 33333 55555 22222 22222 44444 44444 11111 33333 33333 55555 22222 22222 44444 44444 11111 33333 33333 55555 22222 22222 44444 44444 11111 33333 33333 55555 22222 22222 44444 44444
Numeric values are deprecated but are accepted for backward compatibility reasons.
Default mode is "merge".
Available value for flags is:
Vertical low-pass filtering and bypassing already interlaced frames can only be enabled for mode interleave_top and interleave_bottom.
Pick median pixels from several successive input video frames.
The filter accepts the following options:
Mix successive video frames.
A description of the accepted options follows.
Examples
tmix=frames=7:weights="1 1 1 1 1 1 1"
tmix=frames=3:weights="-1 3 -1"
tmix=frames=3:weights="-1 2 -1":scale=1
Tone map colors from different dynamic ranges.
This filter expects data in single precision floating point, as it needs to operate on (and can output) out-of-range values. Another filter, such as zscale, is needed to convert the resulting frame to a usable format.
The tonemapping algorithms implemented only work on linear light, so input data should be linearized beforehand (and possibly correctly tagged).
ffmpeg -i INPUT -vf zscale=transfer=linear,tonemap=clip,zscale=transfer=bt709,format=yuv420p OUTPUT
Options
The filter accepts the following options.
Possible values are:
Default is none.
This affects the following algorithms:
The default of 2.0 is somewhat conservative and will mostly just apply to skies or directly sunlit surfaces. A setting of 0.0 disables this option.
This option works only if the input frame has a supported color tag.
Temporarily pad video frames.
The filter accepts the following options:
The default value of color is "black".
Transpose rows with columns in the input video and optionally flip it.
It accepts the following parameters:
Can assume the following values:
L.R L.l . . -> . . l.r R.r
L.R l.L . . -> . . l.r r.R
L.R R.r . . -> . . l.r L.l
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.
Default value is "none".
For example to rotate by 90 degrees clockwise and preserve portrait layout:
transpose=dir=1:passthrough=portrait
The command above can also be specified as:
transpose=1:portrait
Transpose rows with columns in the input video and optionally flip it. For more in depth examples see the transpose video filter, which shares mostly the same options.
It accepts the following parameters:
Can assume the following values:
Trim the input so that the output contains one continuous subpart of the input.
It accepts the following parameters:
start, end, and duration are expressed as time duration specifications; see the Time duration section in the ffmpeg-utils(1) manual for the accepted syntax.
Note that the first two sets of the start/end options and the duration option look at the frame timestamp, while the _frame variants simply count the frames that pass through the filter. Also note that this filter does not modify the timestamps. If you wish for the output timestamps to start at zero, insert a setpts filter after the trim filter.
If multiple start or end options are set, this filter tries to be greedy and keep all the frames that match at least one of the specified constraints. To keep only the part that matches all the constraints at once, chain multiple trim filters.
The defaults are such that all the input is kept. So it is possible to set e.g. just the end values to keep everything before the specified time.
Examples:
ffmpeg -i INPUT -vf trim=60:120
ffmpeg -i INPUT -vf trim=duration=1
Apply alpha unpremultiply effect to input video stream using first plane of second stream as alpha.
Both streams must have same dimensions and same pixel format.
The filter accepts the following option:
If the format has 1 or 2 components, then luma is bit 0. If the format has 3 or 4 components: for RGB formats bit 0 is green, bit 1 is blue and bit 2 is red; for YUV formats bit 0 is luma, bit 1 is chroma-U and bit 2 is chroma-V. If present, the alpha channel is always the last bit.
Sharpen or blur the input video.
It accepts the following parameters:
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.
Negative values will blur the input video, while positive values will sharpen it, a value of zero will disable the effect.
Default value is 0.0.
All parameters are optional and default to the equivalent of the string '5:5:1.0:5:5:0.0'.
Examples
unsharp=luma_msize_x=7:luma_msize_y=7:luma_amount=2.5
unsharp=7:7:-2:7:7:-2
Decompose a video made of tiled images into the individual images.
The frame rate of the output video is the frame rate of the input video multiplied by the number of tiles.
This filter does the reverse of tile.
The filter accepts the following options:
Examples
ffmpeg -r 1 -i image.jpg -vf untile=1x25 movie.mkv
Apply ultra slow/simple postprocessing filter that compresses and decompresses the image at several (or - in the case of quality level 8 - all) shifts and average the results.
The way this differs from the behavior of spp is that uspp actually encodes & decodes each case with libavcodec Snow, whereas spp uses a simplified intra only 8x8 DCT similar to MJPEG.
The filter accepts the following options:
Convert 360 videos between various formats.
The filter accepts the following options:
Format specific options:
Example values:
Default value is @samp{0}. Maximum value is @samp{0.1}.
Default value is @samp{0}. If greater than zero it overrides other padding options.
Designation of directions:
Default value is @samp{rludfb}.
Designation of angles:
Default value is @samp{000000}.
Format specific options:
If diagonal field of view is set it overrides horizontal and vertical field of view.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Format specific options:
If diagonal field of view is set it overrides horizontal and vertical field of view.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Format specific options:
If diagonal field of view is set it overrides horizontal and vertical field of view.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Format specific options:
If diagonal field of view is set it overrides horizontal and vertical field of view.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Format specific options:
Format specific options:
If diagonal field of view is set it overrides horizontal and vertical field of view.
If diagonal field of view is set it overrides horizontal and vertical field of view.
Format specific options:
Available methods:
Default value is @samp{line}.
Default resolution depends on formats.
Default value is @samp{2d} for input and output format.
Default value is @samp{ypr}.
Examples
ffmpeg -i input.mkv -vf v360=e:c3x2:cubic:out_pad=0.01 output.mkv
ffmpeg -i input.mkv -vf v360=eac:flat:yaw=180 output.mkv
v360=eac:equirect:in_stereo=sbs:in_trans=1:ih_flip=1:out_stereo=tb
Commands
This filter supports subset of above options as commands.
Apply a wavelet based denoiser.
It transforms each frame from the video input into the wavelet domain, using Cohen-Daubechies-Feauveau 9/7. Then it applies some filtering to the obtained coefficients. It does an inverse wavelet transform after. Due to wavelet properties, it should give a nice smoothed result, and reduced noise, without blurring picture features.
This filter accepts the following options:
It accepts the following values:
Default is garrote.
It accepts the following values:
Default is universal.
Display 2 color component values in the two dimensional graph (which is called a vectorscope).
This filter accepts the following options:
It accepts the following values:
Default is auto.
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:
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.
Examples
vidstabdetect
vidstabdetect=shakiness=10:accuracy=15:result="mytransforms.trf"
vidstabdetect=show=1
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".
Options
For example a number of 10 means that 21 frames are used (10 in the past and 10 in the future) to smoothen the motion in the video. A larger value leads to a smoother video, but limits the acceleration of the camera (pan/tilt movements). 0 is a special case where a static camera is simulated.
Accepted values are:
Available values are:
Accepted values are:
Note that the value given at zoom is added to the one calculated here.
Available values are:
Use also "tripod" option of vidstabdetect.
Examples
ffmpeg -i inp.mpeg -vf vidstabtransform,unsharp=5:5:0.8:3:3:0.4 inp_stabilized.mpeg
Note the use of the unsharp filter which is always recommended.
vidstabtransform=zoom=5:input="mytransforms.trf"
vidstabtransform=smoothing=30
Flip the input video vertically.
For example, to vertically flip a video with ffmpeg:
ffmpeg -i in.avi -vf "vflip" out.avi
Detect variable frame rate video.
This filter tries to detect if the input is variable or constant frame rate.
At end it will output number of frames detected as having variable delta pts, and ones with constant delta pts. If there was frames with variable delta, than it will also show min, max and average delta encountered.
Boost or alter saturation.
The filter accepts the following options:
Commands
This filter supports the all above options as commands.
Make or reverse a natural vignetting effect.
The filter accepts the following options:
The value is clipped in the "[0,PI/2]" range.
Default value: "PI/5"
Available modes are:
Default value is forward.
It accepts the following values:
Default value is init.
Default is "1/1".
Expressions
The alpha, x0 and y0 expressions can contain the following parameters.
Examples
vignette=PI/4
vignette='PI/4+random(1)*PI/50':eval=frame
Obtain the average VMAF motion score of a video. It is one of the component metrics of VMAF.
The obtained average motion score is printed through the logging system.
The filter accepts the following options:
Example:
ffmpeg -i ref.mpg -vf vmafmotion -f null -
Stack input videos vertically.
All streams must be of same pixel format and of same width.
Note that this filter is faster than using overlay and pad filter to create same output.
The filter accepts the following options:
Deinterlace the input video ("w3fdif" stands for "Weston 3 Field Deinterlacing Filter").
Based on the process described by Martin Weston for BBC R&D, and implemented based on the de-interlace algorithm written by Jim Easterbrook for BBC R&D, the Weston 3 field deinterlacing filter uses filter coefficients calculated by BBC R&D.
This filter uses field-dominance information in frame to decide which of each pair of fields to place first in the output. If it gets it wrong use setfield filter before "w3fdif" filter.
There are two sets of filter coefficients, so called "simple" and "complex". Which set of filter coefficients is used can be set by passing an optional parameter:
Default value is complex.
Default value is all.
Video waveform monitor.
The waveform monitor plots color component intensity. By default luminance only. Each column of the waveform corresponds to a column of pixels in the source video.
It accepts the following options:
This display mode makes it easier to spot relative differences or similarities in overlapping areas of the color components that are supposed to be identical, such as neutral whites, grays, or blacks.
Using this display mode makes it easy to spot color casts in the highlights and shadows of an image, by comparing the contours of the top and the bottom graphs of each waveform. Since whites, grays, and blacks are characterized by exactly equal amounts of red, green, and blue, neutral areas of the picture should display three waveforms of roughly equal width/height. If not, the correction is easy to perform by making level adjustments the three waveforms.
Default is "stack".
Default is digital.
The "weave" takes a field-based video input and join each two sequential fields into single frame, producing a new double height clip with half the frame rate and half the frame count.
The "doubleweave" works same as "weave" but without halving frame rate and frame count.
It accepts the following option:
Examples
separatefields,select=eq(mod(n,4),0)+eq(mod(n,4),3),weave
Apply the xBR high-quality magnification filter which is designed for pixel art. It follows a set of edge-detection rules, see <https://forums.libretro.com/t/xbr-algorithm-tutorial/123>.
It accepts the following option:
Apply cross fade from one input video stream to another input video stream. The cross fade is applied for specified duration.
The filter accepts the following options:
Default transition effect is fade.
The expressions can use the following variables and functions:
Examples
ffmpeg -i first.mp4 -i second.mp4 -filter_complex xfade=transition=fade:duration=2:offset=5 output.mp4
Pick median pixels from several input videos.
The filter accepts the following options:
Stack video inputs into custom layout.
All streams must be of same pixel format.
The filter accepts the following options:
Note that if inputs are of different sizes gaps may appear, as not all of the output video frame will be filled. Similarly, videos can overlap each other if their position doesn't leave enough space for the full frame of adjoining videos.
For 2 inputs, a default layout of "0_0|w0_0" is set. In all other cases, a layout must be set by the user.
Examples
Layout:
input1(0, 0) | input3(w0, 0) input2(0, h0) | input4(w0, h0) xstack=inputs=4:layout=0_0|0_h0|w0_0|w0_h0
Note that if inputs are of different sizes, gaps or overlaps may occur.
Layout:
input1(0, 0) input2(0, h0) input3(0, h0+h1) input4(0, h0+h1+h2) xstack=inputs=4:layout=0_0|0_h0|0_h0+h1|0_h0+h1+h2
Note that if inputs are of different widths, unused space will appear.
Layout:
input1(0, 0) | input4(w0, 0) | input7(w0+w3, 0) input2(0, h0) | input5(w0, h0) | input8(w0+w3, h0) input3(0, h0+h1) | input6(w0, h0+h1) | input9(w0+w3, h0+h1) xstack=inputs=9:layout=0_0|0_h0|0_h0+h1|w0_0|w0_h0|w0_h0+h1|w0+w3_0|w0+w3_h0|w0+w3_h0+h1
Note that if inputs are of different sizes, gaps or overlaps may occur.
Layout:
input1(0, 0) | input5(w0, 0) | input9 (w0+w4, 0) | input13(w0+w4+w8, 0) input2(0, h0) | input6(w0, h0) | input10(w0+w4, h0) | input14(w0+w4+w8, h0) input3(0, h0+h1) | input7(w0, h0+h1) | input11(w0+w4, h0+h1) | input15(w0+w4+w8, h0+h1) input4(0, h0+h1+h2)| input8(w0, h0+h1+h2)| input12(w0+w4, h0+h1+h2)| input16(w0+w4+w8, h0+h1+h2) xstack=inputs=16:layout=0_0|0_h0|0_h0+h1|0_h0+h1+h2|w0_0|w0_h0|w0_h0+h1|w0_h0+h1+h2|w0+w4_0| w0+w4_h0|w0+w4_h0+h1|w0+w4_h0+h1+h2|w0+w4+w8_0|w0+w4+w8_h0|w0+w4+w8_h0+h1|w0+w4+w8_h0+h1+h2
Note that if inputs are of different sizes, gaps or overlaps may occur.
Deinterlace the input video ("yadif" means "yet another deinterlacing filter").
It accepts the following parameters:
The default value is "send_frame".
The default value is "auto". If the interlacing is unknown or the decoder does not export this information, top field first will be assumed.
The default value is "all".
Deinterlace the input video using the yadif algorithm, but implemented in CUDA so that it can work as part of a GPU accelerated pipeline with nvdec and/or nvenc.
It accepts the following parameters:
The default value is "send_frame".
The default value is "auto". If the interlacing is unknown or the decoder does not export this information, top field first will be assumed.
The default value is "all".
Apply blur filter while preserving edges ("yaepblur" means "yet another edge preserving blur filter"). The algorithm is described in "J. S. Lee, Digital image enhancement and noise filtering by use of local statistics, IEEE Trans. Pattern Anal. Mach. Intell. PAMI-2, 1980."
It accepts the following parameters:
Commands
This filter supports same commands as options.
Apply Zoom & Pan effect.
This filter accepts the following options:
Each expression can contain the following constants:
Examples
zoompan=z='min(zoom+0.0015,1.5)':d=700:x='if(gte(zoom,1.5),x,x+1/a)':y='if(gte(zoom,1.5),y,y+1)':s=640x360
zoompan=z='min(zoom+0.0015,1.5)':d=700:x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)'
zoompan=z='min(max(zoom,pzoom)+0.0015,1.5)':d=1:x='iw/2-(iw/zoom/2)':y='ih/2-(ih/zoom/2)'
Scale (resize) the input video, using the z.lib library: <https://github.com/sekrit-twc/zimg>. To enable compilation of this filter, you need to configure FFmpeg with "--enable-libzimg".
The zscale filter forces the output display aspect ratio to be the same as the input, by changing the output sample aspect ratio.
If the input image format is different from the format requested by the next filter, the zscale filter will convert the input to the requested format.
Options
The filter accepts the following options.
If the width or w value is 0, the input width is used for the output. If the height or h value is 0, the input height is used for the output.
If one and only one of the values is -n with n >= 1, the zscale filter will use a value that maintains the aspect ratio of the input image, calculated from the other specified dimension. After that it will, however, make sure that the calculated dimension is divisible by n and adjust the value if necessary.
If both values are -n with n >= 1, the behavior will be identical to both values being set to 0 as previously detailed.
See below for the list of accepted constants for use in the dimension expression.
Possible values are:
Default is none.
Possible values are:
Default is bilinear.
Possible values are:
Default is same as input.
Possible values are:
Default is same as input.
Possible values are:
Default is same as input.
Possible value are:
Default is same as input.
Possible values are:
Default is same as input.
Possible values are:
Default is same as input.
Possible values are:
Default is same as input.
Possible value are:
Possible values are:
Possible values are:
The values of the w and h options are expressions containing the following constants:
Commands
This filter supports the following commands:
Below is a description of the currently available OpenCL video filters.
To enable compilation of these filters you need to configure FFmpeg with "--enable-opencl".
Running OpenCL filters requires you to initialize a hardware device and to pass that device to all filters in any filter graph.
For more detailed information see <https://www.ffmpeg.org/ffmpeg.html#Advanced-Video-options>
-init_hw_device opencl=gpu:1.0 -filter_hw_device gpu -i INPUT -vf "hwupload, avgblur_opencl, hwdownload" OUTPUT
Since OpenCL filters are not able to access frame data in normal memory, all frame data needs to be uploaded(hwupload) to hardware surfaces connected to the appropriate device before being used and then downloaded(hwdownload) back to normal memory. Note that hwupload will upload to a surface with the same layout as the software frame, so it may be necessary to add a format filter immediately before to get the input into the right format and hwdownload does not support all formats on the output - it may be necessary to insert an additional format filter immediately following in the graph to get the output in a supported format.
Apply average blur filter.
The filter accepts the following options:
Example
-i INPUT -vf "hwupload, avgblur_opencl=3, hwdownload" OUTPUT
Apply a boxblur algorithm to the input video.
It accepts the following parameters:
A description of the accepted options follows.
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:
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.
Examples
Apply boxblur filter, setting each pixel of the output to the average value of box-radiuses luma_radius, chroma_radius, alpha_radius for each plane respectively. The filter will apply luma_power, chroma_power, alpha_power times onto the corresponding plane. For pixels on the edges of the image, the radius does not extend beyond the image boundaries, and so out-of-range coordinates are not used in the calculations.
-i INPUT -vf "hwupload, boxblur_opencl=luma_radius=2:luma_power=3, hwdownload" OUTPUT -i INPUT -vf "hwupload, boxblur_opencl=2:3, hwdownload" OUTPUT
For the luma plane, a 2x2 box radius will be run once.
For the chroma plane, a 4x4 box radius will be run 5 times.
For the alpha plane, a 3x3 box radius will be run 7 times.
-i INPUT -vf "hwupload, boxblur_opencl=2:1:4:5:3:7, hwdownload" OUTPUT
RGB colorspace color keying.
The filter accepts the following options:
0.01 matches only the exact key color, while 1.0 matches everything.
0.0 makes pixels either fully transparent, or not transparent at all.
Higher values result in semi-transparent pixels, with a higher transparency the more similar the pixels color is to the key color.
Examples
-i INPUT -vf "hwupload, colorkey_opencl=green:0.3:0.1, hwdownload" OUTPUT
Apply convolution of 3x3, 5x5, 7x7 matrix.
The filter accepts the following options:
Examples
-i INPUT -vf "hwupload, convolution_opencl=0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0:0 -1 0 -1 5 -1 0 -1 0, hwdownload" OUTPUT
-i INPUT -vf "hwupload, convolution_opencl=1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1 1 1 1 1 1 1 1 1:1/9:1/9:1/9:1/9, hwdownload" OUTPUT
-i INPUT -vf "hwupload, convolution_opencl=0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:0 0 0 -1 1 0 0 0 0:5:1:1:1:0:128:128:128, hwdownload" OUTPUT
-i INPUT -vf "hwupload, convolution_opencl=0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:0 1 0 1 -4 1 0 1 0:5:5:5:1:0:128:128:128, hwdownload" OUTPUT
-i INPUT -vf "hwupload, convolution_opencl=1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:1 1 1 1 -8 1 1 1 1:5:5:5:1:0:128:128:0, hwdownload" OUTPUT
-i INPUT -vf "hwupload, convolution_opencl=-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2:-2 -1 0 -1 1 1 0 1 2, hwdownload" OUTPUT
Apply erosion effect to the video.
This filter replaces the pixel by the local(3x3) minimum.
It accepts the following options:
Flags to local 3x3 coordinates region centered on "x":
1 2 3 4 x 5 6 7 8
Example
-i INPUT -vf "hwupload, erosion_opencl=30:40:50:coordinates=231, hwdownload" OUTPUT
Feature-point based video stabilization filter.
The filter accepts the following options:
Note that in order to see console debug output you will also need to pass "-v verbose" to ffmpeg.
Viewing point matches in the output video is only supported for RGB input.
Defaults to 0.
Defaults to 1.
This can be turned off for a slight performance gain at the cost of precision.
Defaults to 1.
1.0 is the maximum smoothing strength while values less than that result in less smoothing.
0.0 causes the filter to adaptively choose a smoothing strength on a per-frame basis.
Defaults to 0.0.
The size of the smoothing window is determined by multiplying the framerate of the video by this number.
Acceptable values range from 0.1 to 10.0.
Larger values increase the amount of motion data available for determining how to smooth the camera path, potentially improving smoothness, but also increase latency and memory usage.
Defaults to 2.0.
Examples
-i INPUT -vf "hwupload, deshake_opencl=smooth_strength=0.5, hwdownload" OUTPUT
-i INPUT -filter_complex "[0:v]format=rgba, hwupload, deshake_opencl=debug=1, hwdownload, format=rgba, format=yuv420p" -v verbose OUTPUT
Apply dilation effect to the video.
This filter replaces the pixel by the local(3x3) maximum.
It accepts the following options:
Flags to local 3x3 coordinates region centered on "x":
1 2 3 4 x 5 6 7 8
Example
-i INPUT -vf "hwupload, dilation_opencl=30:40:50:coordinates=231, hwdownload" OUTPUT
Non-local Means denoise filter through OpenCL, this filter accepts same options as nlmeans.
Overlay one video on top of another.
It takes two inputs and has one output. The first input is the "main" video on which the second input is overlaid. This filter requires same memory layout for all the inputs. So, format conversion may be needed.
The filter accepts the following options:
Examples
-i INPUT -i LOGO -filter_complex "[0:v]hwupload[a], [1:v]format=yuv420p, hwupload[b], [a][b]overlay_opencl, hwdownload" OUTPUT
-i INPUT -i LOGO -filter_complex "[0:v]hwupload[a], [1:v]format=yuva420p, hwupload[b], [a][b]overlay_opencl, hwdownload" OUTPUT
Add paddings to the input image, and place the original input at the provided x, y coordinates.
It accepts the following options:
The width expression can reference the value set by the height expression, and vice versa.
The default value of width and height is 0.
The x expression can reference the value set by the y expression, and vice versa.
The default value of x and y is 0.
If x or y evaluate to a negative number, they'll be changed so the input image is centered on the padded area.
The value for the width, height, x, and y options are expressions containing the following constants:
Apply the Prewitt operator (<https://en.wikipedia.org/wiki/Prewitt_operator>) to input video stream.
The filter accepts the following option:
Example
-i INPUT -vf "hwupload, prewitt_opencl=scale=2:delta=10, hwdownload" OUTPUT
Filter video using an OpenCL program.
The "program_opencl" filter also supports the framesync options.
The program source file must contain a kernel function with the given name, which will be run once for each plane of the output. Each run on a plane gets enqueued as a separate 2D global NDRange with one work-item for each pixel to be generated. The global ID offset for each work-item is therefore the coordinates of a pixel in the destination image.
The kernel function needs to take the following arguments:
This image will become the output; the kernel should write all of it.
This is a counter starting from zero and increasing by one for each frame.
These are the most recent images on each input. The kernel may read from them to generate the output, but they can't be written to.
Example programs:
__kernel void copy(__write_only image2d_t destination, unsigned int index, __read_only image2d_t source) { const sampler_t sampler = CLK_NORMALIZED_COORDS_FALSE; int2 location = (int2)(get_global_id(0), get_global_id(1)); float4 value = read_imagef(source, sampler, location); write_imagef(destination, location, value); }
__kernel void rotate_image(__write_only image2d_t dst, unsigned int index, __read_only image2d_t src) { const sampler_t sampler = (CLK_NORMALIZED_COORDS_FALSE | CLK_FILTER_LINEAR); float angle = (float)index / 100.0f; float2 dst_dim = convert_float2(get_image_dim(dst)); float2 src_dim = convert_float2(get_image_dim(src)); float2 dst_cen = dst_dim / 2.0f; float2 src_cen = src_dim / 2.0f; int2 dst_loc = (int2)(get_global_id(0), get_global_id(1)); float2 dst_pos = convert_float2(dst_loc) - dst_cen; float2 src_pos = { cos(angle) * dst_pos.x - sin(angle) * dst_pos.y, sin(angle) * dst_pos.x + cos(angle) * dst_pos.y }; src_pos = src_pos * src_dim / dst_dim; float2 src_loc = src_pos + src_cen; if (src_loc.x < 0.0f || src_loc.y < 0.0f || src_loc.x > src_dim.x || src_loc.y > src_dim.y) write_imagef(dst, dst_loc, 0.5f); else write_imagef(dst, dst_loc, read_imagef(src, sampler, src_loc)); }
__kernel void blend_images(__write_only image2d_t dst, unsigned int index, __read_only image2d_t src1, __read_only image2d_t src2) { const sampler_t sampler = (CLK_NORMALIZED_COORDS_FALSE | CLK_FILTER_LINEAR); float blend = (cos((float)index / 50.0f) + 1.0f) / 2.0f; int2 dst_loc = (int2)(get_global_id(0), get_global_id(1)); int2 src1_loc = dst_loc * get_image_dim(src1) / get_image_dim(dst); int2 src2_loc = dst_loc * get_image_dim(src2) / get_image_dim(dst); float4 val1 = read_imagef(src1, sampler, src1_loc); float4 val2 = read_imagef(src2, sampler, src2_loc); write_imagef(dst, dst_loc, val1 * blend + val2 * (1.0f - blend)); }
Apply the Roberts cross operator (<https://en.wikipedia.org/wiki/Roberts_cross>) to input video stream.
The filter accepts the following option:
Example
-i INPUT -vf "hwupload, roberts_opencl=scale=2:delta=10, hwdownload" OUTPUT
Apply the Sobel operator (<https://en.wikipedia.org/wiki/Sobel_operator>) to input video stream.
The filter accepts the following option:
Example
-i INPUT -vf "hwupload, sobel_opencl=scale=2:delta=10, hwdownload" OUTPUT
Perform HDR(PQ/HLG) to SDR conversion with tone-mapping.
It accepts the following parameters:
The default value is 0.5, and the algorithm here is a little different from the cpu version tonemap currently. A setting of 0.0 disables this option.
Currently supported formats are:
Possible values are:
Default is same as input.
Possible values are:
Default is same as input.
Possible values are:
Default is bt709.
Possible value are:
Default is same as input.
Example
-i INPUT -vf "format=p010,hwupload,tonemap_opencl=t=bt2020:tonemap=linear:format=p010,hwdownload,format=p010" OUTPUT
Sharpen or blur the input video.
It accepts the following parameters:
Negative values will blur the input video, while positive values will sharpen it, a value of zero will disable the effect.
Negative values will blur the input video, while positive values will sharpen it, a value of zero will disable the effect.
All parameters are optional and default to the equivalent of the string '5:5:1.0:5:5:0.0'.
Examples
-i INPUT -vf "hwupload, unsharp_opencl=luma_msize_x=7:luma_msize_y=7:luma_amount=2.5, hwdownload" OUTPUT
-i INPUT -vf "hwupload, unsharp_opencl=7:7:-2:7:7:-2, hwdownload" OUTPUT
Cross fade two videos with custom transition effect by using OpenCL.
It accepts the following options:
The program source file must contain a kernel function with the given name, which will be run once for each plane of the output. Each run on a plane gets enqueued as a separate 2D global NDRange with one work-item for each pixel to be generated. The global ID offset for each work-item is therefore the coordinates of a pixel in the destination image.
The kernel function needs to take the following arguments:
This image will become the output; the kernel should write all of it.
These are the most recent images on each input. The kernel may read from them to generate the output, but they can't be written to.
Example programs:
__kernel void blend_images(__write_only image2d_t dst, __read_only image2d_t src1, __read_only image2d_t src2, float progress) { const sampler_t sampler = (CLK_NORMALIZED_COORDS_FALSE | CLK_FILTER_LINEAR); int2 p = (int2)(get_global_id(0), get_global_id(1)); float2 rp = (float2)(get_global_id(0), get_global_id(1)); float2 dim = (float2)(get_image_dim(src1).x, get_image_dim(src1).y); rp = rp / dim; float2 dots = (float2)(20.0, 20.0); float2 center = (float2)(0,0); float2 unused; float4 val1 = read_imagef(src1, sampler, p); float4 val2 = read_imagef(src2, sampler, p); bool next = distance(fract(rp * dots, &unused), (float2)(0.5, 0.5)) < (progress / distance(rp, center)); write_imagef(dst, p, next ? val1 : val2); }
VAAPI Video filters are usually used with VAAPI decoder and VAAPI encoder. Below is a description of VAAPI video filters.
To enable compilation of these filters you need to configure FFmpeg with "--enable-vaapi".
To use vaapi filters, you need to setup the vaapi device correctly. For more information, please read <https://trac.ffmpeg.org/wiki/Hardware/VAAPI>
Perform HDR(High Dynamic Range) to SDR(Standard Dynamic Range) conversion with tone-mapping. It maps the dynamic range of HDR10 content to the SDR content. It currently only accepts HDR10 as input.
It accepts the following parameters:
Currently supported formats are:
Default is nv12.
Default is same as input.
Default is bt709.
Default is same as input.
Example
tonemap_vaapi=format=p010:t=bt2020-10
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.
It accepts the following parameters:
For example:
buffer=width=320:height=240:pix_fmt=yuv410p:time_base=1/24:sar=1
will instruct the source to accept video frames with size 320x240 and with format "yuv410p", assuming 1/24 as the timestamps timebase and square pixels (1:1 sample aspect ratio). Since the pixel format with name "yuv410p" corresponds to the number 6 (check the enum AVPixelFormat definition in libavutil/pixfmt.h), this example corresponds to:
buffer=size=320x240:pixfmt=6:time_base=1/24:pixel_aspect=1/1
Alternatively, the options can be specified as a flat string, but this syntax is deprecated:
width:height:pix_fmt:time_base.num:time_base.den:pixel_aspect.num:pixel_aspect.den
Create a pattern generated by an elementary cellular automaton.
The initial state of the cellular automaton can be defined through the filename and pattern options. If such options are not specified an initial state is created randomly.
At each new frame a new row in the video is filled with the result of the cellular automaton next generation. The behavior when the whole frame is filled is defined by the scroll option.
This source accepts the following options:
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.
This option is ignored when a file or a pattern is specified.
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).
Examples
cellauto=f=pattern:s=200x400
cellauto=ratio=2/3:s=200x200
cellauto=p=@s=100x400:full=0:rule=18
cellauto=p='@@ @ @@':s=100x400:full=0:rule=18
Video source generated on GPU using Apple's CoreImage API on OSX.
This video source is a specialized version of the coreimage video filter. Use a core image generator at the beginning of the applied filterchain to generate the content.
The coreimagesrc video source accepts the following options:
list_generators=true
If not specified, or the expressed duration is negative, the video is supposed to be generated forever.
Additionally, all options of the coreimage video filter are accepted. A complete filterchain can be used for further processing of the generated input without CPU-HOST transfer. See coreimage documentation and examples for details.
Examples
ffmpeg -f lavfi -i coreimagesrc=s=100x100:filter=CIQRCodeGenerator@inputMessage=https\\\\\://FFmpeg.org/@inputCorrectionLevel=H -frames:v 1 QRCode.png
This example is equivalent to the QRCode example of coreimage without the need for a nullsrc video source.
Generate several gradients.
Generate a Mandelbrot set fractal, and progressively zoom towards the point specified with start_x and start_y.
This source accepts the following options:
It shall assume one of the following values:
Default value is mincol.
Default value is normalized_iteration_count.
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:
If not specified, or the expressed duration is negative, the video is supposed to be generated forever.
Default value is "all", which will cycle through the list of all tests.
Some examples:
mptestsrc=t=dc_luma
will generate a "dc_luma" test pattern.
Provide a frei0r source.
To enable compilation of this filter you need to install the frei0r header and configure FFmpeg with "--enable-frei0r".
This source accepts the following parameters:
For example, to generate a frei0r partik0l source with size 200x200 and frame rate 10 which is overlaid on the overlay filter main input:
frei0r_src=size=200x200:framerate=10:filter_name=partik0l:filter_params=1234 [overlay]; [in][overlay] overlay
Generate a life pattern.
This source is based on a generalization of John Conway's life game.
The sourced input represents a life grid, each pixel represents a cell which can be in one of two possible states, alive or dead. Every cell interacts with its eight neighbours, which are the cells that are horizontally, vertically, or diagonally adjacent.
At each interaction the grid evolves according to the adopted rule, which specifies the number of neighbor alive cells which will make a cell stay alive or born. The rule option allows one to specify the rule to adopt.
This source accepts the following options:
If this option is not specified, the initial grid is generated randomly.
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.
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).
For the syntax of these 3 color options, check the "Color" section in the ffmpeg-utils manual.
Examples
life=f=pattern:s=300x300
life=ratio=2/3:s=200x200
life=rule=S14/B34
ffplay -f lavfi life=s=300x200:mold=10:r=60:ratio=0.1:death_color=#C83232:life_color=#00ff00,scale=1200:800:flags=16
The "allrgb" source returns frames of size 4096x4096 of all rgb colors.
The "allyuv" source returns frames of size 4096x4096 of all yuv colors.
The "color" source provides an uniformly colored input.
The "haldclutsrc" source provides an identity Hald CLUT. See also haldclut filter.
The "nullsrc" source returns unprocessed video frames. It is mainly useful to be employed in analysis / debugging tools, or as the source for filters which ignore the input data.
The "pal75bars" source generates a color bars pattern, based on EBU PAL recommendations with 75% color levels.
The "pal100bars" source generates a color bars pattern, based on EBU PAL recommendations with 100% color levels.
The "rgbtestsrc" source generates an RGB test pattern useful for detecting RGB vs BGR issues. You should see a red, green and blue stripe from top to bottom.
The "smptebars" source generates a color bars pattern, based on the SMPTE Engineering Guideline EG 1-1990.
The "smptehdbars" source generates a color bars pattern, based on the SMPTE RP 219-2002.
The "testsrc" source generates a test video pattern, showing a color pattern, a scrolling gradient and a timestamp. This is mainly intended for testing purposes.
The "testsrc2" source is similar to testsrc, but supports more pixel formats instead of just "rgb24". This allows using it as an input for other tests without requiring a format conversion.
The "yuvtestsrc" source generates an YUV test pattern. You should see a y, cb and cr stripe from top to bottom.
The sources accept the following parameters:
This option is not available with the "allrgb", "allyuv", and "haldclutsrc" filters.
If not specified, or the expressed duration is negative, the video is supposed to be generated forever.
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.
Examples
testsrc=duration=5.3:size=qcif:rate=10
color=c=red@0.2:s=qcif:r=10
nullsrc=s=256x256, geq=random(1)*255:128:128
Commands
The "color" source supports the following commands:
Generate video using an OpenCL program.
For details of how the program loading works, see the program_opencl filter.
Example programs:
__kernel void ramp(__write_only image2d_t dst, unsigned int index) { int2 loc = (int2)(get_global_id(0), get_global_id(1)); float4 val; val.xy = val.zw = convert_float2(loc) / convert_float2(get_image_dim(dst)); write_imagef(dst, loc, val); }
__kernel void sierpinski_carpet(__write_only image2d_t dst, unsigned int index) { int2 loc = (int2)(get_global_id(0), get_global_id(1)); float4 value = 0.0f; int x = loc.x + index; int y = loc.y + index; while (x > 0 || y > 0) { if (x % 3 == 1 && y % 3 == 1) { value = 1.0f; break; } x /= 3; y /= 3; } write_imagef(dst, loc, value); }
Generate a Sierpinski carpet/triangle fractal, and randomly pan around.
This source accepts the following options:
Below is a description of the currently available video sinks.
Buffer video frames, and make them available to the end of the filter graph.
This sink is mainly intended for programmatic use, in particular through the interface defined in libavfilter/buffersink.h or the options system.
It accepts a pointer to an AVBufferSinkContext structure, which defines the incoming buffers' formats, to be passed as the opaque parameter to "avfilter_init_filter" for initialization.
Null video sink: do absolutely nothing with the input video. It is mainly useful as a template and for use in analysis / debugging tools.
Below is a description of the currently available multimedia filters.
Convert input audio to a video output, displaying the audio bit scope.
The filter accepts the following options:
Draw a graph using input audio metadata.
See drawgraph
See graphmonitor.
Convert input audio to a video output, displaying the volume histogram.
The filter accepts the following options:
It accepts the following values:
Default is "single".
It accepts the following values:
Default is "log".
It accepts the following values:
Default is "log".
It accepts the following values:
Default is "replace".
Measures phase of input audio, which is exported as metadata "lavfi.aphasemeter.phase", representing mean phase of current audio frame. A video output can also be produced and is enabled by default. The audio is passed through as first output.
Audio will be rematrixed to stereo if it has a different channel layout. Phase value is in range "[-1, 1]" where "-1" means left and right channels are completely out of phase and 1 means channels are in phase.
The filter accepts the following options, all related to its video output:
Convert input audio to a video output, representing the audio vector scope.
The filter is used to measure the difference between channels of stereo audio stream. A monaural signal, consisting of identical left and right signal, results in straight vertical line. Any stereo separation is visible as a deviation from this line, creating a Lissajous figure. If the straight (or deviation from it) but horizontal line appears this indicates that the left and right channels are out of phase.
The filter accepts the following options:
Available values are:
Default value is lissajous.
Available values are:
Default value is dot.
Available values are:
Examples
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]'
Benchmark part of a filtergraph.
The filter accepts the following options:
Available values are:
Examples
bench=start,selectivecolor=reds=-.2 .12 -.49,bench=stop
Concatenate audio and video streams, joining them together one after the other.
The filter works on segments of synchronized video and audio streams. All segments must have the same number of streams of each type, and that will also be the number of streams at output.
The filter accepts the following options:
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.
Examples
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.
Commands
This filter supports the following commands:
EBU R128 scanner filter. This filter takes an audio stream and analyzes its loudness level. By default, it logs a message at a frequency of 10Hz with the Momentary loudness (identified by "M"), Short-term loudness ("S"), Integrated loudness ("I") and Loudness Range ("LRA").
The filter can only analyze streams which have a sampling rate of 48000 Hz and whose sample format is double-precision floating point. The input stream will be converted to this specification, if needed. Users may need to insert aformat and/or aresample filters after this filter to obtain the original parameters.
The filter also has a video output (see the video option) with a real time graph to observe the loudness evolution. The graphic contains the logged message mentioned above, so it is not printed anymore when this option is set, unless the verbose logging is set. The main graphing area contains the short-term loudness (3 seconds of analysis), and the gauge on the right is for the momentary loudness (400 milliseconds), but can optionally be configured to instead display short-term loudness (see gauge).
The green area marks a +/- 1LU target range around the target loudness (-23LUFS by default, unless modified through target).
More information about the Loudness Recommendation EBU R128 on <http://tech.ebu.ch/loudness>.
The filter accepts the following options:
Default is 0.
Available values are:
By default, the logging level is set to info. If the video or the metadata options are set, it switches to verbose.
Available modes can be cumulated (the option is a "flag" type). Possible values are:
Simple peak mode looking for the higher sample value. It logs a message for sample-peak (identified by "SPK").
If enabled, the peak lookup is done on an over-sampled version of the input stream for better peak accuracy. It logs a message for true-peak. (identified by "TPK") and true-peak per frame (identified by "FTPK"). This mode requires a build with "libswresample".
Examples
ffplay -f lavfi -i "amovie=input.mp3,ebur128=video=1:meter=18 [out0][out1]"
ffmpeg -nostats -i input.mp3 -filter_complex ebur128 -f null -
Temporally interleave frames from several inputs.
"interleave" works with video inputs, "ainterleave" with audio.
These filters read frames from several inputs and send the oldest queued frame to the output.
Input streams must have well defined, monotonically increasing frame timestamp values.
In order to submit one frame to output, these filters need to enqueue at least one frame for each input, so they cannot work in case one input is not yet terminated and will not receive incoming frames.
For example consider the case when one input is a "select" filter which always drops input frames. The "interleave" filter will keep reading from that input, but it will never be able to send new frames to output until the input sends an end-of-stream signal.
Also, depending on inputs synchronization, the filters will drop frames in case one input receives more frames than the other ones, and the queue is already filled.
These filters accept the following options:
Examples
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
Manipulate frame metadata.
This filter accepts the following options:
Can be one of the following:
Can be one of following:
Examples
signalstats,metadata=print:key=lavfi.signalstats.YDIF:value=0:function=expr:expr='between(VALUE1,0,1)'
silencedetect,ametadata=mode=print:file=metadata.txt
metadata=mode=print:file='pipe\:4'
Set read/write permissions for the output frames.
These filters are mainly aimed at developers to test direct path in the following filter in the filtergraph.
The filters accept the following options:
It accepts the following values:
Note: in case of auto-inserted filter between the permission filter and the following one, the permission might not be received as expected in that following filter. Inserting a format or aformat filter before the perms/aperms filter can avoid this problem.
Slow down filtering to match real time approximately.
These filters will pause the filtering for a variable amount of time to match the output rate with the input timestamps. They are similar to the re option to "ffmpeg".
They accept the following options:
A processing speed faster than what is possible without these filters cannot be achieved.
Select frames to pass in output.
This filter accepts the following options:
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.
The expression can contain the following constants:
This works by comparing the frame pts against the lavf.concat.start_time and the lavf.concat.duration packet metadata values which are also present in the decoded frames.
The concatdec_select variable is -1 if the frame pts is at least start_time and either the duration metadata is missing or the frame pts is less than start_time + duration, 0 otherwise, and NaN if the start_time metadata is missing.
That basically means that an input frame is selected if its pts is within the interval set by the concat demuxer.
The default value of the select expression is "1".
Examples
select
The example above is the same as:
select=1
select=0
select='eq(pict_type\,I)'
select='not(mod(n\,100))'
select=between(t\,10\,20)
select=between(t\,10\,20)*eq(pict_type\,I)
select='isnan(prev_selected_t)+gte(t-prev_selected_t\,10)'
aselect='gt(samples_n\,100)'
ffmpeg -i video.avi -vf select='gt(scene\,0.4)',scale=160:120,tile -frames:v 1 preview.png
Comparing scene against a value between 0.3 and 0.5 is generally a sane choice.
select=n=2:e='mod(n, 2)+1' [odd][even]; [odd] pad=h=2*ih [tmp]; [tmp][even] overlay=y=h
ffmpeg -copyts -vsync 0 -segment_time_metadata 1 -i input.ffconcat -vf select=concatdec_select -af aselect=concatdec_select output.avi
Send commands to filters in the filtergraph.
These filters read commands to be sent to other filters in the filtergraph.
"sendcmd" must be inserted between two video filters, "asendcmd" must be inserted between two audio filters, but apart from that they act the same way.
The specification of commands can be provided in the filter arguments with the commands option, or in a file specified by the filename option.
These filters accept the following options:
Commands syntax
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 expression is evaluated through the eval API and can contain the following constants:
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>]
Examples
asendcmd=c='4.0 atempo tempo 1.5',atempo
asendcmd=c='4.0 atempo@my tempo 1.5',atempo@my
# show text in the interval 5-10 5.0-10.0 [enter] drawtext reinit 'fontfile=FreeSerif.ttf:text=hello world', [leave] drawtext reinit 'fontfile=FreeSerif.ttf:text='; # desaturate the image in the interval 15-20 15.0-20.0 [enter] hue s 0, [enter] drawtext reinit 'fontfile=FreeSerif.ttf:text=nocolor', [leave] hue s 1, [leave] drawtext reinit 'fontfile=FreeSerif.ttf:text=color'; # apply an exponential saturation fade-out effect, starting from time 25 25 [enter] hue s exp(25-t)
A filtergraph allowing to read and process the above command list stored in a file test.cmd, can be specified with:
sendcmd=f=test.cmd,drawtext=fontfile=FreeSerif.ttf:text='',hue
Change the PTS (presentation timestamp) of the input frames.
"setpts" works on video frames, "asetpts" on audio frames.
This filter accepts the following options:
The expression is evaluated through the eval API and can contain the following constants:
Examples
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
Force color range for the output video frame.
The "setrange" filter marks the color range property for the output frames. It does not change the input frame, but only sets the corresponding property, which affects how the frame is treated by following filters.
The filter accepts the following options:
Set the timebase to use for the output frames timestamps. It is mainly useful for testing timebase configuration.
It accepts the following parameters:
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".
Examples
settb=expr=1/25
settb=expr=0.1
settb=1+0.001
settb=2*intb
settb=AVTB
Convert input audio to a video output representing frequency spectrum logarithmically using Brown-Puckette constant Q transform algorithm with direct frequency domain coefficient calculation (but the transform itself is not really constant Q, instead the Q factor is actually variable/clamped), with musical tone scale, from E0 to D#10.
The filter accepts the following options:
and functions:
Default value is 16.
and functions:
Default value is "sono_v".
Default value is "384*tc/(384+tc*f)".
and functions:
Default value is "st(0, (midi(f)-59.5)/12); st(1, if(between(ld(0),0,1), 0.5-0.5*cos(2*PI*ld(0)), 0)); r(1-ld(1)) + b(ld(1))".
Examples
ffplay -f lavfi 'amovie=a.mp3, asplit [a][out1]; [a] showcqt [out0]'
ffplay -f lavfi 'amovie=a.mp3, asplit [a][out1]; [a] showcqt=fps=30:count=5 [out0]'
ffplay -f lavfi 'amovie=a.mp3, asplit [a][out1]; [a] showcqt=s=1280x720:count=4 [out0]'
sono_h=0
ffplay -f lavfi 'aevalsrc=0.1*sin(2*PI*55*t)+0.1*sin(4*PI*55*t)+0.1*sin(6*PI*55*t)+0.1*sin(8*PI*55*t), asplit[a][out1]; [a] showcqt [out0]'
ffplay -f lavfi 'aevalsrc=0.1*sin(2*PI*55*t)+0.1*sin(4*PI*55*t)+0.1*sin(6*PI*55*t)+0.1*sin(8*PI*55*t), asplit[a][out1]; [a] showcqt=timeclamp=0.5 [out0]'
bar_v=10:sono_v=bar_v*a_weighting(f)
bar_g=2:sono_g=2
tc=0.33:tlength='st(0,0.17); 384*tc / (384 / ld(0) + tc*f /(1-ld(0))) + 384*tc / (tc*f / ld(0) + 384 /(1-ld(0)))'
fontcolor='if(mod(floor(midi(f)+0.5),12), 0x0000FF, g(1))':fontfile=myfont.ttf
font='Courier New,Monospace,mono|bold'
axisfile=myaxis.png:basefreq=40:endfreq=10000
Convert input audio to video output representing the audio power spectrum. Audio amplitude is on Y-axis while frequency is on X-axis.
The filter accepts the following options:
It accepts the following values:
Default is "bar".
It accepts the following values:
Default is "log".
It accepts the following values:
Default is "lin".
Default is 2048
It accepts the following values:
Default is "hanning".
It accepts the following values:
Default is "combined".
Convert stereo input audio to a video output, representing the spatial relationship between two channels.
The filter accepts the following options:
It accepts the following values:
Default value is "hann".
Convert input audio to a video output, representing the audio frequency spectrum.
The filter accepts the following options:
It accepts the following values:
Default value is "replace".
It accepts the following values:
Default value is combined.
It accepts the following values:
Default value is channel.
It accepts the following values:
Default value is sqrt.
It accepts the following values:
Default value is lin.
It accepts the following values:
Default value is "hann".
The usage is very similar to the showwaves filter; see the examples in that section.
Examples
showspectrum=s=1280x480:scale=log
ffplay -f lavfi 'amovie=input.mp3, asplit [a][out1]; [a] showspectrum=mode=separate:color=intensity:slide=1:scale=cbrt [out0]'
Convert input audio to a single video frame, representing the audio frequency spectrum.
The filter accepts the following options:
It accepts the following values:
Default value is combined.
It accepts the following values:
Default value is intensity.
It accepts the following values:
Default value is log.
It accepts the following values:
Default value is lin.
It accepts the following values:
Default value is "hann".
Examples
ffmpeg -i audio.flac -lavfi showspectrumpic=s=1024x1024 spectrogram.png
Convert input audio volume to a video output.
The filter accepts the following options:
The expression can use the following variables:
Convert input audio to a video output, representing the samples waves.
The filter accepts the following options:
Available values are:
Default value is "point".
Available values are:
Default is linear.
Available values are:
Default value is "scale".
Examples
amovie=a.mp3,asplit[out0],showwaves[out1]
aevalsrc=sin(1*2*PI*t)*sin(880*2*PI*t):cos(2*PI*200*t),asplit[out0],showwaves=r=30[out1]
Convert input audio to a single video frame, representing the samples waves.
The filter accepts the following options:
Available values are:
Default is linear.
Available values are:
Default value is "scale".
Examples
ffmpeg -i audio.flac -lavfi showwavespic=split_channels=1:s=1024x800 waveform.png
Delete frame side data, or select frames based on it.
This filter accepts the following options:
Can be one of the following:
Synthesize audio from 2 input video spectrums, first input stream represents magnitude across time and second represents phase across time. The filter will transform from frequency domain as displayed in videos back to time domain as presented in audio output.
This filter is primarily created for reversing processed showspectrum filter outputs, but can synthesize sound from other spectrograms too. But in such case results are going to be poor if the phase data is not available, because in such cases phase data need to be recreated, usually it's just recreated from random noise. For best results use gray only output ("channel" color mode in showspectrum filter) and "log" scale for magnitude video and "lin" scale for phase video. To produce phase, for 2nd video, use "data" option. Inputs videos should generally use "fullframe" slide mode as that saves resources needed for decoding video.
The filter accepts the following options:
Examples
ffmpeg -i input.flac -lavfi showspectrum=mode=separate:scale=log:overlap=0.875:color=channel:slide=fullframe:data=magnitude -an -c:v rawvideo magnitude.nut ffmpeg -i input.flac -lavfi showspectrum=mode=separate:scale=lin:overlap=0.875:color=channel:slide=fullframe:data=phase -an -c:v rawvideo phase.nut ffmpeg -i magnitude.nut -i phase.nut -lavfi spectrumsynth=channels=2:sample_rate=44100:win_func=hann:overlap=0.875:slide=fullframe output.flac
Split input into several identical outputs.
"asplit" works with audio input, "split" with video.
The filter accepts a single parameter which specifies the number of outputs. If unspecified, it defaults to 2.
Examples
[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 -i INPUT -filter_complex asplit=5 OUTPUT
Receive commands sent through a libzmq client, and forward them to filters in the filtergraph.
"zmq" and "azmq" work as a pass-through filters. "zmq" must be inserted between two video filters, "azmq" between two audio filters. Both are capable to send messages to any filter type.
To enable these filters you need to install the libzmq library and headers and configure FFmpeg with "--enable-libzmq".
For more information about libzmq see: <http://www.zeromq.org/>
The "zmq" and "azmq" filters work as a libzmq server, which receives messages sent through a network interface defined by the bind_address (or the abbreviation "b") option. Default value of this option is tcp://localhost:5555. You may want to alter this value to your needs, but do not forget to escape any ':' signs (see filtergraph escaping).
The received message must be in the form:
<TARGET> <COMMAND> [<ARG>]
TARGET specifies the target of the command, usually the name of the filter class or a specific filter instance name. The default filter instance name uses the pattern Parsed_<filter_name>_<index>, but you can override this by using the filter_name@id syntax (see Filtergraph syntax).
COMMAND specifies the name of the command for the target filter.
ARG is optional and specifies the optional argument list for the given COMMAND.
Upon reception, the message is processed and the corresponding command is injected into the filtergraph. Depending on the result, the filter will send a reply to the client, adopting the format:
<ERROR_CODE> <ERROR_REASON> <MESSAGE>
MESSAGE is optional.
Examples
Look at tools/zmqsend for an example of a zmq client which can be used to send commands processed by these filters.
Consider the following filtergraph generated by ffplay. In this example the last overlay filter has an instance name. All other filters will have default instance names.
ffplay -dumpgraph 1 -f lavfi " color=s=100x100:c=red [l]; color=s=100x100:c=blue [r]; nullsrc=s=200x100, zmq [bg]; [bg][l] overlay [bg+l]; [bg+l][r] overlay@my=x=100 "
To change the color of the left side of the video, the following command can be used:
echo Parsed_color_0 c yellow | tools/zmqsend
To change the right side:
echo Parsed_color_1 c pink | tools/zmqsend
To change the position of the right side:
echo overlay@my x 150 | tools/zmqsend
Below is a description of the currently available multimedia sources.
This is the same as movie source, except it selects an audio stream by default.
Read audio and/or video stream(s) from a movie container.
It accepts the following parameters:
Note that when the movie is looped the source timestamps are not changed, so it will generate non monotonically increasing timestamps.
It allows overlaying a second video on top of the main input of a filtergraph, as shown in this graph:
input -----------> deltapts0 --> overlay --> output ^ | movie --> scale--> deltapts1 -------+
Examples
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]
Commands
Both movie and amovie support the following commands:
The FFmpeg developers.
For details about the authorship, see the Git history of the project (https://git.ffmpeg.org/ffmpeg), e.g. by typing the command git log in the FFmpeg source directory, or browsing the online repository at <https://git.ffmpeg.org/ffmpeg>.
Maintainers for the specific components are listed in the file MAINTAINERS in the source code tree.