FFMPEG(1)
💅 FFMPEG(1) — the ffmpeg media converter, bestie edition ✨
Okay so this is the FFMPEG(1) manual, and we are gonna make it make sense. No cap.
🎀 NAME 🎀
ffmpeg — ffmpeg media converter. That’s the whole name section, bestie. Slay.
✨ SYNOPSIS ✨
ffmpeg [global_options] {[input_file_options] -i input_url} ...
{[output_file_options] output_url} ...🦋 DESCRIPTION 🦋
So ffmpeg is literally a universal media converter. It can read a wide variety of inputs — including live grabbing/recording devices — filter them, and transcode them into a plethora of output formats. Obsessed.
ffmpeg reads from an arbitrary number of inputs (which can be regular files, pipes, network streams, grabbing devices, etc.), specified by the -i option, and writes to an arbitrary number of outputs, which are specified by a plain output url. Anything found on the command line which cannot be interpreted as an option is considered to be an output url. That’s the vibe check.
Each input or output can, in principle, contain any number of elementary streams of different types (video/audio/subtitle/attachment/data), though the allowed stream counts and/or types may be limited by the container format. Selecting which streams from which inputs will go into which output is either done automatically or with the -map option (see the Stream selection chapter).
To refer to inputs/outputs in options, you must use their indices (0-based). E.g. the first input is 0, the second is 1, etc. Similarly, streams within an input/output are referred to by their indices. E.g. 2:3 refers to the fourth stream in the third input or output. Also see the Stream specifiers chapter. Iykyk.
As a general rule, options are applied to the next specified file. Therefore, order is important, and you can have the same option on the command line multiple times. Each occurrence is then applied to the next input or output file. Exceptions from this rule are the global options (e.g. verbosity level), which should be specified first.
🚨 Do not mix input and output files — first specify all input files, then all output files. Also do not mix options which belong to different files. All options apply ONLY to the next input or output file and are reset between files. Periodt.
Some simple examples follow.
- Convert an input media file to a different format, by re-encoding media streams:
ffmpeg -i input.avi output.mp4- Set the video bitrate of the output file to 64 kbit/s:
ffmpeg -i input.avi -b:v 64k -bufsize 64k output.mp4- Force the frame rate of the output file to 24 fps:
ffmpeg -i input.avi -r 24 output.mp4- Force the frame rate of the input file (valid for raw formats only) to 1 fps and the frame rate of the output file to 24 fps:
ffmpeg -r 1 -i input.m2v -r 24 output.mp4The format option may be needed for raw input files. Don’t forget that, bestie.
🌈 DETAILED DESCRIPTION 🌈
ffmpeg builds a transcoding pipeline out of the components listed below. The program’s operation then consists of input data chunks flowing from the sources down the pipes towards the sinks, while being transformed by the components they encounter along the way. It’s giving assembly line, but make it media.
The following kinds of components are available:
- Demuxers (short for “demultiplexers”) — read an input source in order to extract:
- global properties such as metadata or chapters;
- list of input elementary streams and their properties.
One demuxer instance is created for each -i option, and sends encoded packets to decoders or muxers.
In other literature, demuxers are sometimes called splitters, because their main function is splitting a file into elementary streams (though some files only contain one elementary stream).
A schematic representation of a demuxer looks like this:
┌──────────┬───────────────────────┐
│ demuxer │ │ packets for stream 0
╞══════════╡ elementary stream 0 ├──────────────────────►
│ │ │
│ global ├───────────────────────┤
│properties│ │ packets for stream 1
│ and │ elementary stream 1 ├──────────────────────►
│ metadata │ │
│ ├───────────────────────┤
│ │ │
│ │ ........... │
│ │ │
│ ├───────────────────────┤
│ │ │ packets for stream N
│ │ elementary stream N ├──────────────────────►
│ │ │
└──────────┴───────────────────────┘
▲
│
│ read from file, network stream,
│ grabbing device, etc.
│- Decoders — receive encoded (compressed) packets for an audio, video, or subtitle elementary stream, and decode them into raw frames (arrays of pixels for video, PCM for audio). A decoder is typically associated with (and receives its input from) an elementary stream in a demuxer, but sometimes may also exist on its own (see Loopback decoders).
A schematic representation of a decoder looks like this:
┌─────────┐
packets │ │ raw frames
─────────►│ decoder ├────────────►
│ │
└─────────┘- Filtergraphs — process and transform raw audio or video frames. A filtergraph consists of one or more individual filters linked into a graph. Filtergraphs come in two flavors — simple and complex, configured with the
-filterand-filter_complexoptions, respectively.
A simple filtergraph is associated with an output elementary stream; it receives the input to be filtered from a decoder and sends filtered output to that output stream’s encoder.
A simple video filtergraph that performs deinterlacing (using the yadif deinterlacer) followed by resizing (using the scale filter) can look like this:
┌────────────────────────┐
│ simple filtergraph │
frames from ╞════════════════════════╡ frames for
a decoder │ ┌───────┐ ┌───────┐ │ an encoder
────────────►├─►│ yadif ├─►│ scale ├─►│────────────►
│ └───────┘ └───────┘ │
└────────────────────────┘A complex filtergraph is standalone and not associated with any specific stream. It may have multiple (or zero) inputs, potentially of different types (audio or video), each of which receiving data either from a decoder or another complex filtergraph’s output. It also has one or more outputs that feed either an encoder or another complex filtergraph’s input. Main character energy, honestly.
The following example diagram represents a complex filtergraph with 3 inputs and 2 outputs (all video):
┌─────────────────────────────────────────────────┐
│ complex filtergraph │
╞═════════════════════════════════════════════════╡
frames ├───────┐ ┌─────────┐ ┌─────────┐ ┌────────┤ frames
─────────►│input 0├─►│ overlay ├─────►│ overlay ├─►│output 0├────────►
├───────┘ │ │ │ │ └────────┤
frames ├───────┐╭►│ │ ╭►│ │ │
─────────►│input 1├╯ └─────────┘ │ └─────────┘ │
├───────┘ │ │
frames ├───────┐ ┌─────┐ ┌─────┬─╯ ┌────────┤ frames
─────────►│input 2├►│scale├►│split├───────────────►│output 1├────────►
├───────┘ └─────┘ └─────┘ └────────┤
└─────────────────────────────────────────────────┘Frames from second input are overlaid over those from the first. Frames from the third input are rescaled, then the duplicated into two identical streams. One of them is overlaid over the combined first two inputs, with the result exposed as the filtergraph’s first output. The other duplicate ends up being the filtergraph’s second output. Iconic behavior.
- Encoders — receive raw audio, video, or subtitle frames and encode them into encoded packets. The encoding (compression) process is typically lossy — it degrades stream quality to make the output smaller; some encoders are lossless, but at the cost of much higher output size. A video or audio encoder receives its input from some filtergraph’s output, subtitle encoders receive input from a decoder (since subtitle filtering is not supported yet). Every encoder is associated with some muxer’s output elementary stream and sends its output to that muxer.
A schematic representation of an encoder looks like this:
So the encoder hands off like this, bestie:
┌─────────┐
raw frames │ │ packets
────────────►│ encoder ├─────────►
│ │
└─────────┘- Muxers (short for "multiplexers") — okay so the muxers are the ones receiving encoded packets for their elementary streams, and they get them from encoders (that's the transcoding path 🛣️) or straight-up directly from demuxers (that's the streamcopy path). Then they interleave them (only when there's more than one elementary stream, obvi) and write the resulting bytes into the output file (or pipe, network stream, etc.). A schematic of a muxer is literally giving:
┌──────────────────────┬───────────┐
packets for stream 0 │ │ muxer │
──────────────────────►│ elementary stream 0 ╞═══════════╡
│ │ │
├──────────────────────┤ global │
packets for stream 1 │ │properties │
──────────────────────►│ elementary stream 1 │ and │
│ │ metadata │
├──────────────────────┤ │
│ │ │
│ ........... │ │
│ │ │
├──────────────────────┤ │
packets for stream N │ │ │
──────────────────────►│ elementary stream N │ │
│ │ │
└──────────────────────┴─────┬─────┘
│
write to file, network stream, │
grabbing device, etc. │
│
▼🦋 Streamcopy 🦋
Okay so the simplest pipeline in ffmpeg is single-stream streamcopy — that's copying one input elementary stream's packets without decoding, filtering, or encoding them. Zero extra steps, we love a low-effort icon 💅
Take an input file called INPUT.mkv with 3 elementary streams, and we're grabbing the second one and writing it to OUTPUT.mp4. The schematic of that pipeline is:
┌──────────┬─────────────────────┐
│ demuxer │ │ unused
╞══════════╡ elementary stream 0 ├────────╳
│ │ │
│INPUT.mkv ├─────────────────────┤ ┌──────────────────────┬───────────┐
│ │ │ packets │ │ muxer │
│ │ elementary stream 1 ├─────────►│ elementary stream 0 ╞═══════════╡
│ │ │ │ │OUTPUT.mp4 │
│ ├─────────────────────┤ └──────────────────────┴───────────┘
│ │ │ unused
│ │ elementary stream 2 ├────────╳
│ │ │
└──────────┴─────────────────────┘And you build that whole vibe with this commandline:
ffmpeg -i INPUT.mkv -map 0:1 -c copy OUTPUT.mp4In this commandline:
- there is a single input
INPUT.mkv; - there are no input options for this input;
- there is a single output
OUTPUT.mp4; - there are two output options for this output:
-map 0:1selects the input stream to be used — from input with index 0 (i.e. the first one) the stream with index 1 (i.e. the second one);-c copyselects thecopyencoder, i.e. streamcopy with no decoding or encoding.
Streamcopy is lowkey the goat for changing the elementary stream count, container format, or modifying container-level metadata. Since there's no decoding or encoding happening, it's very fast and there's no quality loss. Periodt. ✨
But bestie, hear me — it might not work in some cases because of a variety of factors (e.g. certain information required by the target container is not available in the source). Applying filters is obviously also impossible, since filters work on decoded frames. No decoded frames = no filter slay. 😭
More complex streamcopy scenarios can be constructed too — like combining streams from two input files into a single output:
┌──────────┬────────────────────┐ ┌────────────────────┬───────────┐
│ demuxer 0│ │ packets │ │ muxer │
╞══════════╡elementary stream 0 ├────────►│elementary stream 0 ╞═══════════╡
│INPUT0.mkv│ │ │ │OUTPUT.mp4 │
└──────────┴────────────────────┘ ├────────────────────┤ │
┌──────────┬────────────────────┐ │ │ │
│ demuxer 1│ │ packets │elementary stream 1 │ │
╞══════════╡elementary stream 0 ├────────►│ │ │
│INPUT1.aac│ │ └────────────────────┴───────────┘
└──────────┴────────────────────┘which can be built by the commandline
ffmpeg -i INPUT0.mkv -i INPUT1.aac -map 0:0 -map 1:0 -c copy OUTPUT.mp4The output -map option is used twice here, creating two streams in the output file — one fed by the first input and one by the second. The single instance of the -c option selects streamcopy for both of those streams. You could also use multiple instances of this option together with Stream specifiers to apply different values to each stream, as will be demonstrated in following sections. 👀
A converse scenario is splitting multiple streams from a single input into multiple outputs:
┌──────────┬─────────────────────┐ ┌───────────────────┬───────────┐
│ demuxer │ │ packets │ │ muxer 0 │
╞══════════╡ elementary stream 0 ├─────────►│elementary stream 0╞═══════════╡
│ │ │ │ │OUTPUT0.mp4│
│INPUT.mkv ├─────────────────────┤ └───────────────────┴───────────┘
│ │ │ packets ┌───────────────────┬───────────┐
│ │ elementary stream 1 ├─────────►│ │ muxer 1 │
│ │ │ │elementary stream 0╞═══════════╡
└──────────┴─────────────────────┘ │ │OUTPUT1.mp4│
└───────────────────┴───────────┘built with
ffmpeg -i INPUT.mkv -map 0:0 -c copy OUTPUT0.mp4 -map 0:1 -c copy OUTPUT1.mp4Not me needing to point out: a separate instance of the -c option is needed for every output file, even though their values are the same. Why? Because non-global options (which is most of them) only apply in the context of the file before which they are placed. Positions matter, bestie. 🎯
These examples can of course be further generalized into arbitrary remappings of any number of inputs into any number of outputs. Go feral. 🌀
🌈 Transcoding 🌈
Transcoding is the process of decoding a stream and then encoding it again. And since encoding tends to be computationally expensive and in most cases degrades the stream quality (i.e. it is lossy), you should only transcode when you need to and perform streamcopy otherwise. Do not transcode for fun, I'm begging 🙏
Typical reasons to transcode are:
- applying filters — e.g. resizing, deinterlacing, or overlaying video; resampling or mixing audio;
- you want to feed the stream to something that cannot decode the original codec.
Heads up:ffmpegwill transcode all audio, video, and subtitle streams unless you specify-c copyfor them. So if you only meant to touch one thing, say so, or the whole file is getting re-encoded. 😭
Consider an example pipeline that reads an input file with one audio and one video stream, transcodes the video and copies the audio into a single output file. This can be schematically represented as follows
okay so here’s the full pipeline picture for that exact transcoding flow — yes, it’s giving very much diagram energy 💅
┌──────────┬─────────────────────┐
│ demuxer │ │ audio packets
╞══════════╡ stream 0 (audio) ├─────────────────────────────────────╮
│ │ │ │
│INPUT.mkv ├─────────────────────┤ video ┌─────────┐ raw │
│ │ │ packets │ video │ video frames │
│ │ stream 1 (video) ├─────────►│ decoder ├──────────────╮ │
│ │ │ │ │ │ │
└──────────┴─────────────────────┘ └─────────┘ │ │
▼ ▼
│ │
┌──────────┬─────────────────────┐ video ┌─────────┐ │ │
│ muxer │ │ packets │ video │ │ │
╞══════════╡ stream 0 (video) │◄─────────┤ encoder ├──────────────╯ │
│ │ │ │(libx264)│ │
│OUTPUT.mp4├─────────────────────┤ └─────────┘ │
│ │ │ │
│ │ stream 1 (audio) │◄────────────────────────────────────╯
│ │ │
└──────────┴─────────────────────┘and the commandline that implements it is literally:
ffmpeg -i INPUT.mkv -map 0:v -map 0:a -c:v libx264 -c:a copy OUTPUT.mp4Notice how it uses stream specifiers :v and :a to select input streams and apply different values of the -c option to them; see the Stream specifiers section for more details. 🫶
Filtering ✨
When you’re transcoding, audio and video streams can be filtered before encoding, with either a simple or complex filtergraph. no cap, that’s the whole vibe check.
Simple filtergraphs 🎀
Simple filtergraphs are those that have exactly one input and output, both of the same type (audio or video). They’re configured with the per-stream -filter option (with -vf and -af aliases for -filter:v (video) and -filter:a (audio) respectively). Heads up: simple filtergraphs are tied to their output stream, so e.g. if you have multiple audio streams, -af will create a separate filtergraph for each one. the way that works is lowkey important 💅
Taking the transcoding example from above, adding filtering (and omitting audio, for clarity) makes it look like this:
┌──────────┬───────────────┐
│ demuxer │ │ ┌─────────┐
╞══════════╡ video stream │ packets │ video │ frames
│INPUT.mkv │ ├─────────►│ decoder ├─────►───╮
│ │ │ └─────────┘ │
└──────────┴───────────────┘ │
╭───────────◄───────────╯
│ ┌────────────────────────┐
│ │ simple filtergraph │
│ ╞════════════════════════╡
│ │ ┌───────┐ ┌───────┐ │
╰──►├─►│ yadif ├─►│ scale ├─►├╮
│ └───────┘ └───────┘ ││
└────────────────────────┘│
│
│
┌──────────┬───────────────┐ video ┌─────────┐ │
│ muxer │ │ packets │ video │ │
╞══════════╡ video stream │◄─────────┤ encoder ├───────◄───────╯
│OUTPUT.mp4│ │ │ │
│ │ │ └─────────┘
└──────────┴───────────────┘Complex filtergraphs 🌈
Complex filtergraphs are those which cannot be described as simply a linear processing chain applied to one stream. This is the case, for example, when the graph has more than one input and/or output, or when output stream type is different from input. Complex filtergraphs are configured with the -filter_complex option. Note that this option is global, since a complex filtergraph, by its nature, cannot be unambiguously associated with a single stream or file. Each instance of -filter_complex creates a new complex filtergraph, and there can be any number of them. periodt.
A trivial example of a complex filtergraph is the overlay filter, which has two video inputs and one video output, containing one video overlaid on top of the other. Its audio counterpart is the amix filter. iconic behavior, honestly.
Loopback decoders 🌀
While decoders are normally associated with demuxer streams, it is also possible to create “loopback” decoders that decode the output from some encoder and allow it to be fed back to complex filtergraphs. This is done with the -dec directive, which takes as a parameter the index of the output stream that should be decoded. Every such directive creates a new loopback decoder, indexed with successive integers starting at zero. These indices should then be used to refer to loopback decoders in complex filtergraph link labels, as described in the documentation for -filter_complex.
Decoding AVOptions can be passed to loopback decoders by placing them before -dec, analogously to input/output options. iykyk 🧚
E.g. the following example:
ffmpeg -i INPUT \
-map 0:v:0 -c:v libx264 -crf 45 -f null - \
-threads 3 -dec 0:0 \
-filter_complex '[0:v][dec:0]hstack[stack]' \
-map '[stack]' -c:v ffv1 OUTPUTreads an input video and
- (line 2) encodes it with
libx264at low quality; - (line 3) decodes this encoded stream using 3 threads;
- (line 4) places decoded video side by side with the original input video;
- (line 5) combined video is then losslessly encoded and written into
OUTPUT.
Such a transcoding pipeline can be represented with the following diagram:
┌──────────┬───────────────┐
│ demuxer │ │ ┌─────────┐ ┌─────────┐ ┌────────────────────┐
╞══════════╡ video stream │ │ video │ │ video │ │ null muxer │
│ INPUT │ ├──►│ decoder ├──┬────────►│ encoder ├─┬─►│(discards its input)│
│ │ │ └─────────┘ │ │(libx264)│ │ └────────────────────┘
└──────────┴───────────────┘ │ └─────────┘ │
╭───────◄──╯ ┌─────────┐ │
│ │loopback │ │
│ ╭─────◄──────┤ decoder ├────◄──╯
│ │ └─────────┘
│ │
│ │
│ │ ┌───────────────────┐
│ │ │complex filtergraph│
│ │ ╞═══════════════════╡
│ │ │ ┌─────────────┐ │
╰─╫─►├─►│ hstack ├─►├╮
╰─►├─►│ │ ││
│ └─────────────┘ ││
└───────────────────┘│
│
┌──────────┬───────────────┐ ┌─────────┐ │
│ muxer │ │ │ video │ │
╞══════════╡ video stream │◄─┤ encoder ├───────◄──────────╯
│ OUTPUT │ │ │ (ffv1) │
│ │ │ └─────────┘
└──────────┴───────────────┘💅 STREAM SELECTION 💅
okay so ffmpeg gives you the -map option for manual control of stream selection in each output file. You can totally skip -map and let ffmpeg do automatic stream selection like described below. And the -vn / -an / -sn / -dn options can skip inclusion of video, audio, subtitle and data streams respectively — whether they were manually mapped or automatically selected — except for those streams which are outputs of complex filtergraphs. 💖
🫶 Description 🫶
The sub-sections that follow describe the various rules that are involved in stream selection. The examples that follow next show how these rules are applied in practice. ✨
While every effort is made to accurately reflect the behavior of the program, FFmpeg is under continuous development and the code may have changed since the time of this writing. iykyk 👀
✨ Automatic stream selection ✨
In the absence of any map options for a particular output file, ffmpeg inspects the output format to check which type of streams can be included in it, viz. video, audio and/or subtitles. For each acceptable stream type, ffmpeg will pick one stream, when available, from among all the inputs. 🎀
It will select that stream based upon the following criteria:
- for video, it is the stream with the highest resolution,
- for audio, it is the stream with the most channels,
- for subtitles, it is the first subtitle stream found but there's a caveat. The output format's default subtitle encoder can be either text-based or image-based, and only a subtitle stream of the same type will be chosen.
In the case where several streams of the same type rate equally, the stream with the lowest index is chosen. lowkey iconic behavior 👑
Data or attachment streams are not automatically selected and can only be included using -map. No cap. 💀
🎀 Manual stream selection 🎀
When -map is used, only user-mapped streams are included in that output file, with one possible exception for filtergraph outputs described below. 🫶
🦋 Complex filtergraphs 🦋
If there are any complex filtergraph output streams with unlabeled pads, they will be added to the first output file. This will lead to a fatal error if the stream type is not supported by the output format. like, actual fatal error, bestie 😭 In the absence of the map option, the inclusion of these streams leads to the automatic stream selection of their types being skipped. If map options are present, these filtergraph streams are included in addition to the mapped streams.
Complex filtergraph output streams with labeled pads must be mapped once and exactly once. once. and. exactly. once. 💅
🌸 Stream handling 🌸
Stream handling is independent of stream selection, with an exception for subtitles described below. Stream handling is set via the -codec option addressed to streams within a specific output file. In particular, codec options are applied by ffmpeg after the stream selection process and thus do not influence the latter. If no -codec option is specified for a stream type, ffmpeg will select the default encoder registered by the output file muxer. ✨
An exception exists for subtitles. If a subtitle encoder is specified for an output file, the first subtitle stream found of any type, text or image, will be included. ffmpeg does not validate if the specified encoder can convert the selected stream or if the converted stream is acceptable within the output format. This applies generally as well: when the user sets an encoder manually, the stream selection process cannot check if the encoded stream can be muxed into the output file. If it cannot, ffmpeg will abort and all output files will fail to be processed. so maybe double-check that vibe before you hit enter 🙈
💖 Examples 💖
The following examples illustrate the behavior, quirks and limitations of ffmpeg's stream selection methods. 🍓
They assume the following three input files.
input file 'A.avi'
stream 0: video 640x360
stream 1: audio 2 channels
input file 'B.mp4'
stream 0: video 1920x1080
stream 1: audio 2 channels
stream 2: subtitles (text)
stream 3: audio 5.1 channels
stream 4: subtitles (text)
input file 'C.mkv'
stream 0: video 1280x720
stream 1: audio 2 channels
stream 2: subtitles (image)🎬 Example: automatic stream selection 🎬
ffmpeg -i A.avi -i B.mp4 out1.mkv out2.wav -map 1:a -c:a copy out3.movThere are three output files specified, and for the first two, no -map options are set, so ffmpeg will select streams for these two files automatically. 💫
out1.mkv is a Matroska container file and accepts video, audio and subtitle streams, so ffmpeg will try to select one of each type. For video, it will select stream 0 from B.mp4, which has the highest resolution among all the input video streams. For audio, it will select stream 3 from B.mp4, since it has the greatest number of channels. For subtitles, it will select stream 2 from B.mp4, which is the first subtitle stream from among A.avi and B.mp4. slay selection logic 🌈
out2.wav accepts only audio streams, so only stream 3 from B.mp4 is selected. 🎧
For out3.mov, since a -map option is set, no automatic stream selection will occur. The -map 1:a option will select all audio streams from the second input B.mp4. No other streams will be included in this output file. manual mode activated 💅
For the first two outputs, all included streams will be transcoded. The encoders chosen will be the default ones registered by each output format, which may not match the codec of the selected input streams.
For the third output, codec option for audio streams has been set to copy, so no decoding-filtering-encoding operations will occur, or can occur. Packets of selected streams shall be conveyed from the input file and muxed within the output file. copy means copy, no little chemistry set in the middle ✨
🎬 Example: automatic subtitles selection 🎬
ffmpeg -i C.mkv out1.mkv -c:s dvdsub -an out2.mkvAlthough out1.mkv is a Matroska container file which accepts subtitle streams, only a video and audio stream shall be selected. The subtitle stream of C.mkv is image-based and the default subtitle encoder of the Matroska muxer is text-based, so a transcode operation for the subtitles is expected to fail and hence the stream isn't selected. it's giving mismatch 😭 However, in out2.mkv, a subtitle encoder is specified in the command and so, the subtitle stream is selected, in addition to the video stream. The presence of -an disables audio stream selection for out2.mkv. 🎀
🎬 Example: unlabeled filtergraph outputs 🎬
ffmpeg -i A.avi -i C.mkv -i B.mp4 -filter_complex "overlay" out1.mp4 out2.srtA filtergraph is setup here using the -filter_complex option and consists of a single video filter. The overlay filter requires exactly two video inputs, but none are specified, so the first two available video streams are used, those of A.avi and C.mkv. The output pad of the filter has no label and so is sent to the first output file out1.mp4. Due to this, automatic selection of the video stream is skipped, which would have selected the stream in B.mp4. The audio stream with most channels viz. stream 3 in B.mp4, is chosen automatically. No subtitle stream is chosen however, since the MP4 format has no default subtitle encoder registered, and the user hasn't specified a subtitle encoder. 🦋
The 2nd output file, out2.srt, only accepts text-based subtitle streams. So, even though the first subtitle stream available belongs to C.mkv, it is image-based and hence skipped. The selected stream, stream 2 in B.mp4, is the first text-based subtitle stream. first text-based wins, periodt 💅
🎬 Example: labeled filtergraph outputs 🎬
ffmpeg -i A.avi -i B.mp4 -i C.mkv -filter_complex "[1:v]hue=s=0[outv];overlay;aresample" \
-map '[outv]' -an out1.mp4 \
out2.mkv \
-map '[outv]' -map 1:a:0 out3.mkvThe above command will fail, as the output pad labelled [outv] has been mapped twice. None of the output files shall be processed. rip, no outputs for anyone 💀
ffmpeg -i A.avi -i B.mp4 -i C.mkv -filter_complex "[1:v]hue=s=0[outv];overlay;aresample" \
-an out1.mp4 \
out2.mkv \
-map 1:a:0 out3.mkvThis command above will also fail as the hue filter output has a label, [outv], and hasn't been mapped anywhere. labeled pad left on read, and ffmpeg is not okay with that 😭
The command should be modified as follows,
ffmpeg -i A.avi -i B.mp4 -i C.mkv -filter_complex "[1:v]hue=s=0,split=2[outv1][outv2];overlay;aresample" \
-map '[outv1]' -an out1.mp4 \
out2.mkv \
-map '[outv2]' -map 1:a:0 out3.mkvThe video stream from B.mp4 is sent to the hue filter, whose output is cloned once using the split filter, and both outputs labelled. Then a copy each is mapped to the first and third output files. ✨
The overlay filter, requiring two video inputs, uses the first two unused video streams. Those are the streams from A.avi and C.mkv. The overlay output isn't labelled, so it is sent to the first output file out1.mp4, regardless of the presence of the -map option. 🎀
The aresample filter is sent the first unused audio stream, that of A.avi. Since this filter output is also unlabelled, it too is mapped to the first output file. The presence of -an only suppresses automatic or manual stream selection of audio streams, not outputs sent from filtergraphs. Both these mapped streams shall be ordered before the mapped stream in out1.mp4. filtergraph outputs said “we’re not listening to -an” 💅
The video, audio and subtitle streams mapped to out2.mkv are entirely determined by automatic stream selection.
out3.mkv consists of the cloned video output from the hue filter and the first audio stream from B.mp4. iconic little muxing moment 🌈
✨ OPTIONS ✨
All the numerical options, if not specified otherwise, accept a string representing a number as input, which may be followed by one of the SI unit prefixes, for example: K, M, or G. 🔢
If i is appended to the SI unit prefix, the complete prefix will be interpreted as a unit prefix for binary multiples, which are based on powers of 1024 instead of powers of 1000. Appending B to the SI unit prefix multiplies the value by 8. This allows using, for example: KB, MiB, G and B as number suffixes. numbers with little outfits on ✨
Options which do not take arguments are boolean options, and set the corresponding value to true. They can be set to false by prefixing the option name with no. For example using -nofoo will set the boolean option with name foo to false. 🫶
Options that take arguments support a special syntax where the argument given on the command line is interpreted as a path to the file from which the actual argument value is loaded. To use this feature, add a forward slash / immediately before the option name (after the leading dash). E.g.
ffmpeg -i INPUT -/filter:v filter.script OUTPUT...will load a filtergraph description from the file named filter.script. yes bestie, that's literally it 💅
✨ STREAM SPECIFIERS ✨
okay so some options are per-stream — like bitrate or codec, right? and stream specifiers are how you point at exactly which stream(s) an option belongs to 🎯
a stream specifier is a string that you generally slap onto the end of the option name, separated by a colon. like -codec:a:1 ac3 — see the a:1 in there? that's the stream specifier, and it matches the second audio stream. so this command picks the ac3 codec for the second audio stream. slay 💅
and a stream specifier can match several streams at once, so the option just applies to all of them. for example the stream specifier in -b:a 128k matches all audio streams. group effort fr fr 🫶
an empty stream specifier matches all streams. like, -codec copy or -codec: copy would copy all the streams without reencoding. no reencoding, no stress, we love a lazy queen 👑
possible forms of stream specifiers are:
- stream_index — matches the stream with this index. e.g.
-threads:1 4sets the thread count for the second stream to 4. and if stream_index is used as an additional stream specifier (peep below), then it selects stream number stream_index from the matching streams. stream numbering is based on the order of the streams as detected by libavformat — except when a stream group specifier or program ID is also specified, in which case it's based on the ordering of the streams in the group or program. iykyk 🌸 - stream_type[:additional_stream_specifier] — stream_type is one of these:
vorVfor video,afor audio,sfor subtitle,dfor data, andtfor attachments.vmatches all video streams, butVonly matches video streams which are not attached pictures, video thumbnails or cover arts. if additional_stream_specifier is used, then it matches streams which both have this type and match the additional stream specifier. otherwise, it matches all streams of the specified type ✨ - g:group_specifier[:additional_stream_specifier] — matches streams which are in the group with the specifier group_specifier. if additional_stream_specifier is used, then it matches streams which both are part of the group and match the additional_stream_specifier. group_specifier may be one of the following:
- group_index — match the stream with this group index.
#group_idori:group_id— match the stream with this group id.
- p:program_id[:additional_stream_specifier] — matches streams which are in the program with the id program_id. if additional_stream_specifier is used, then it matches streams which both are part of the program and match the additional_stream_specifier 🎀
#stream_idori:stream_id— match the stream by stream id (e.g.PIDin MPEG-TS container). obsessed with the specificity 🦋m:key[:value]— matches streams with the metadata tag key having the specified value. if value is not given, it matches streams that contain the given tag with any value. heads up tho: the colon character:in key or value needs to be backslash-escaped. escape your colons bestie 😭disp:dispositions[:additional_stream_specifier]— matches streams with the given disposition(s). dispositions is a list of one or more dispositions (as printed by the-dispositionsoption) joined with+👑u— matches streams with usable configuration. the codec must be defined and the essential information such as video dimension or audio sample rate must be present. it's giving "vibe check passed" ✅
ngl, note that in ffmpeg, matching by metadata will only work properly for input files. keep that in mind 💀
🌈 GENERIC OPTIONS 🌈
these options are shared amongst the ff* tools. like, all of them. they're family 💖
-L,-license— show license. boring but legally iconic 📜-h,-?,-help,--help [arg]— show help. you can throw an optional parameter in there to print help about a specific item. if no argument is specified, only basic (non advanced) tool options are shown ✨
possible values of arg are:
long— print advanced tool options in addition to the basic tool options.full— print complete list of options, including shared and private options for encoders, decoders, demuxers, muxers, filters, etc. it's literally everything 👑decoder=decoder_name — print detailed information about the decoder named decoder_name. use the-decodersoption to get a list of all decoders.encoder=encoder_name — print detailed information about the encoder named encoder_name. use the-encodersoption to get a list of all encoders.demuxer=demuxer_name — print detailed information about the demuxer named demuxer_name. use the-formatsoption to get a list of all demuxers and muxers.muxer=muxer_name — print detailed information about the muxer named muxer_name. use the-formatsoption to get a list of all muxers and demuxers.filter=filter_name — print detailed information about the filter named filter_name. use the-filtersoption to get a list of all filters.bsf=bitstream_filter_name — print detailed information about the bitstream filter named bitstream_filter_name. use the-bsfsoption to get a list of all bitstream filters.protocol=protocol_name — print detailed information about the protocol named protocol_name. use the-protocolsoption to get a list of all protocols.
-version— show version. say hi to your build 🌸-buildconf— show the build configuration, one option per line. receipts, basically 🧾-formats— show available formats (including devices). the whole lineup 💅-demuxers— show available demuxers.-muxers— show available muxers.-devices— show available devices.-codecs— show all codecs known to libavcodec.
ok important note: the term "codec" is used throughout this documentation as a shortcut for what is more correctly called a media bitstream format. just so we're all on the same page 🫶
-decoders— show available decoders.-encoders— show all available encoders.-bsfs— show available bitstream filters.-protocols— show available protocols.-filters— show available libavfilter filters. yes the library has a filter named libavfilter, we contain multitudes ✨-pix_fmts— show available pixel formats.-sample_fmts— show available sample formats.-layouts— show channel names and standard channel layouts.-dispositions— show stream dispositions.-colors— show recognized color names. aesthetic reference sheet 🎨-sourcesdevice[,opt1=val1[,opt2=val2]...]— show autodetected sources of the input device. some devices may provide system-dependent source names that cannot be autodetected. and the returned list cannot be assumed to be always complete. so like, it's a vibe, not a promise 🙈
ffmpeg -sources pulse,server=192.168.0.4-sinksdevice[,opt1=val1[,opt2=val2]...]— show autodetected sinks of the output device. some devices may provide system-dependent sink names that cannot be autodetected. and same deal, the returned list cannot be assumed to be always complete 💅
ffmpeg -sinks pulse,server=192.168.0.4-loglevel [flags+]loglevel,-v [flags+]loglevel— set logging level and flags used by the library. this is your "how much do you want me to talk" dial 🗣️
the optional flags prefix can consist of the following values:
repeat— indicates that repeated log output should not be compressed to the first line and the "Last message repeated n times" line will be omitted.level— indicates that log output should add a[level]prefix to each message line. this can be used as an alternative to log coloring, e.g. when dumping the log to file.time— indicates that log lines should be prefixed with time information.datetime— indicates that log lines should be prefixed with date and time information. timestamps bestie 🕐
flags can also be used alone by adding a +/- prefix to set/reset a single flag without affecting other flags or changing loglevel. when setting both flags and loglevel, a + separator is expected between the last flags value and before loglevel ✨
loglevel is a string or a number containing one of the following values:
quiet,-8— show nothing at all; be silent. the introvert setting 🤫panic,0— only show fatal errors which could lead the process to crash, such as an assertion failure. this is not currently used for anything. ghost feature fr 💀fatal,8— only show fatal errors. these are errors after which the process absolutely cannot continue.error,16— show all errors, including ones which can be recovered from.warning,24— show all warnings and errors. any message related to possibly incorrect or unexpected events will be shown.info,32— show informative messages during processing. this is in addition to warnings and errors. this is the default value 👑verbose,40— same asinfo, except more verbose. she talks more 💅debug,48— show everything, including debugging information.trace,56— (the maximum of the drama scale, no notes given) 🔥
for example, to enable repeated log output, add the level prefix, and set loglevel to verbose:
ffmpeg -loglevel repeat+level+verbose -i input outputanother example that enables repeated log output without affecting current state of level prefix flag or loglevel:
ffmpeg [...] -loglevel +repeatby default the program logs to stderr. if coloring is supported by the terminal, colors are used to mark errors and warnings. log coloring can be disabled setting the environment variable AV_LOG_FORCE_NOCOLOR, or can be forced setting the environment variable AV_LOG_FORCE_COLOR. pink error messages are a lifestyle 💖
-report— dump full command line and log output to a file namedprogram-YYYYMMDD-HHMMSS.login the current directory. this file can be useful for bug reports. it also implies-loglevel debug📝
setting the environment variable FFREPORT to any value has the same effect. if the value is a :-separated key=value sequence, these options will affect the report; option values must be escaped if they contain special characters or the options delimiter : (see the ``Quoting and escaping'' section in the ffmpeg-utils manual).
the following options are recognized:
file— set the file name to use for the report;%pis expanded to the name of the program,%tis expanded to a timestamp,%%is expanded to a plain%level— set the log verbosity level using a numerical value (see-loglevel).
for example, to output a report to a file named ffreport.log using a log level of 32 (alias for log level info):
FFREPORT=file=ffreport.log:level=32 ffmpeg -i input outputAnd if parsing that environment variable goes sideways? Not fatal, bestie — those errors won’t show up in the report. 🫶
-hide_banner— Suppress printing the banner. All FFmpeg tools will normally show a copyright notice, build options and library versions, and this option can be used to suppress printing this information.-cpuflags flags (global)— Allows setting and clearing cpu flags. This option is intended for testing. Do not use it unless you know what you’re doing.
Examples, because seeing it is believing:
ffmpeg -cpuflags -sse+mmx ...
ffmpeg -cpuflags mmx ...
ffmpeg -cpuflags 0 ...Possible flags for this option are:
- x86
mmxmmxextssesse2sse2slowsse3sse3slowssse3atomsse4.1sse4.2avxavx2xopfma3fma43dnow3dnowextbmi1bmi2cmov
- ARM
armv5tearmv6armv6t2vfpvfpv3neonsetend
- AArch64
armv8vfpneon
- PowerPC
altivec
- Specific Processors
pentium2pentium3pentium4k6k62athlonathlonxpk8
-cpucount count (global)— Override detection of CPU count. This option is intended for testing. Do not use it unless you know what you’re doing.
ffmpeg -cpucount 2-max_alloc bytes— Set the maximum size limit for allocating a block on the heap by ffmpeg’s family of malloc functions. Exercise extreme caution when using this option. Don’t use if you do not understand the full consequence of doing so. Default isINT_MAX.
💖 AVOptions 💖
These options are provided directly by the libavformat, libavdevice and libavcodec libraries. To see the list of available AVOptions, use the -help option. They are separated into two categories:
🌸 generic
These options can be set for any container, codec or device. Generic options are listed under AVFormatContext options for containers/devices and under AVCodecContext options for codecs.
🦋 private
These options are specific to the given container, device or codec. Private options are listed under their corresponding containers/devices/codecs.
For example to write an ID3v2.3 header instead of a default ID3v2.4 to an MP3 file, use the id3v2_version private option of the MP3 muxer:
ffmpeg -i input.flac -id3v2_version 3 out.mp3All codec AVOptions are per-stream, and thus a stream specifier should be attached to them:
ffmpeg -i multichannel.mxf -map 0:v:0 -map 0:a:0 -map 0:a:0 -c:a:0 ac3 -b:a:0 640k -ac:a:1 2 -c:a:1 aac -b:2 128k out.mp4In the above example, a multichannel audio stream is mapped twice for output. The first instance is encoded with codec ac3 and bitrate 640k. The second instance is downmixed to 2 channels and encoded with codec aac. A bitrate of 128k is specified for it using absolute index of the output stream.
⚠️ Note: the-nooptionsyntax cannot be used for boolean AVOptions, use-option 0/-option 1.
⚠️ Note: the old undocumented way of specifying per-stream AVOptions by prepending v/a/s to the options name is now obsolete and will be removed soon.
🌟 Main options 🌟
-f fmt (input/output)— Force input or output file format. The format is normally auto detected for input files and guessed from the file extension for output files, so this option is not needed in most cases.-i url (input)— input file url.-y (global)— Overwrite output files without asking.-n (global)— Do not overwrite output files, and exit immediately if a specified output file already exists.-stream_loop number (input)— Set number of times input stream shall be looped. Loop0means no loop, loop-1means infinite loop.-recast_media (global)— Allow forcing a decoder of a different media type than the one detected or designated by the demuxer. Useful for decoding media data muxed as data streams.-c[:stream_specifier] codec (input/output,per-stream)and-codec[:stream_specifier] codec (input/output,per-stream)— Select an encoder (when used before an output file) or a decoder (when used before an input file) for one or more streams.codecis the name of a decoder/encoder or a special valuecopy(output only) to indicate that the stream is not to be re-encoded.
For example:
ffmpeg -i INPUT -map 0 -c:v libx264 -c:a copy OUTPUTencodes all video streams with libx264 and copies all audio streams.
For each stream, the last matching c option is applied, so:
ffmpeg -i INPUT -map 0 -c copy -c:v:1 libx264 -c:a:137 libvorbis OUTPUTwill copy all the streams except the second video, which will be encoded with libx264, and the 138th audio, which will be encoded with libvorbis.
-t duration (input/output)— When used as an input option (before-i), limit the duration of data read from the input file.
When used as an output option (before an output url), stop writing the output after its duration reaches duration.
duration must be a time duration specification, see the Time duration section in the ffmpeg-utils(1) manual.
-to and -t are mutually exclusive and -t has priority.
-to position (input/output)— Stop writing the output or reading the input atposition.positionmust be a time duration specification, see the Time duration section in theffmpeg-utils(1)manual.
-to and -t are mutually exclusive and -t has priority.
-fs limit_size (output)— Set the file size limit, expressed in bytes. No further chunk of bytes is written after the limit is exceeded. The size of the output file is slightly more than the requested file size.-ss position (input/output)— When used as an input option (before-i), seeks in this input file toposition. Note that in most formats it is not possible to seek exactly, so ffmpeg will seek to the closest seek point beforeposition. When transcoding and-accurate_seekis enabled (the default), this extra segment between the seek point andpositionwill be decoded and discarded. When doing stream copy or when-noaccurate_seekis used, it will be preserved.
When used as an output option (before an output url), decodes but discards input until the timestamps reach position.
position must be a time duration specification, see the Time duration section in the ffmpeg-utils(1) manual.
-sseof position (input)— Like the-ssoption but relative to the"end of file". That is negative values are earlier in the file,0is at EOF.-isync input_index (input)— Assign an input as a sync source.
This will take the difference between the start times of the target and reference inputs and offset the timestamps of the target file by that difference. The source timestamps of the two inputs should derive from the same clock source for expected results. If copyts is set then start_at_zero must also be set. If either of the inputs has no starting timestamp then no sync adjustment is made.
Acceptable values are those that refer to a valid ffmpeg input index. If the sync reference is the target index itself or -1, then no adjustment is made to target timestamps. A sync reference may not itself be synced to any other input.
Default value is -1.
-itsoffset offset (input)— Set the input time offset.
offset must be a time duration specification, see the Time duration section in the ffmpeg-utils(1) manual.
The offset is added to the timestamps of the input files. Specifying a positive offset means that the corresponding streams are delayed by the time duration specified in offset.
-itsscale scale (input,per-stream)— Rescale input timestamps.scaleshould be a floating point number.-timestamp date (output)— Set the recording timestamp in the container.
date must be a date specification, see the Date section in the ffmpeg-utils(1) manual.
-metadata[:metadata_specifier] key=value (output,per-metadata)— Set a metadata key/value pair.
An optional metadata_specifier may be given to set metadata on streams, chapters or programs. See -map_metadata documentation for details.
This option overrides metadata set with -map_metadata. It is also possible to delete metadata by using an empty value.
For example, for setting the title in the output file:
ffmpeg -i in.avi -metadata title="my title" out.flvTo set the language of the first audio stream:
ffmpeg -i INPUT -metadata:s:a:0 language=eng OUTPUT-disposition[:stream_specifier] value (output,per-stream)— Sets the disposition flags for a stream.
Default value: by default, all disposition flags are copied from the input stream, unless the output stream this option applies to is fed by a complex filtergraph — in that case no disposition flags are set by default.
value is a sequence of disposition flags separated by + or -. A + prefix adds the given disposition, - removes it. If the first flag is also prefixed with + or -, the resulting disposition is the default value updated by value. If the first flag is not prefixed, the resulting disposition is value. It is also possible to clear the disposition by setting it to 0.
If no -disposition options were specified for an output file, ffmpeg will automatically set the 'default' disposition flag on the first stream of each type, when there are multiple streams of this type in the output file and no stream of that type is already marked as default.
The -dispositions option lists the known disposition flags.
For example, to make the second audio stream the default stream:
ffmpeg -i in.mkv -c copy -disposition:a:1 default out.mkvTo make the second subtitle stream the default stream and remove the default disposition from the first subtitle stream:
ffmpeg -i in.mkv -c copy -disposition:s:0 0 -disposition:s:1 default out.mkvTo add an embedded cover/thumbnail:
ffmpeg -i in.mp4 -i IMAGE -map 0 -map 1 -c copy -c:v:1 png -disposition:v:1 attached_pic out.mp4okay so, more disposition recipes because flag surgery is a whole art form and we're not done yet 🎀
To add the original flag and yeet the comment flag off the first audio stream — without removing any of its other disposition flags:
ffmpeg -i in.mkv -c copy -disposition:a:0 +original-comment out.mkvTo do the exact opposite — drop original and add comment — again without touching the stream's other disposition flags:
ffmpeg -i in.mkv -c copy -disposition:a:0 -original+comment out.mkvTo set only original and comment on the first audio stream, and wipe all its other disposition flags (no + or - prefix means full replace, no cap, periodt):
ffmpeg -i in.mkv -c copy -disposition:a:0 original+comment out.mkvTo remove every single disposition flag from the first audio stream:
ffmpeg -i in.mkv -c copy -disposition:a:0 0 out.mkv⚠️ real talk: not all muxers support embedded thumbnails, and the ones that do only support a few formats, like JPEG or PNG. So if your container ghosts you, that's why 💀
🎀 -program
[title=title:][program_num=program_num:]st=stream[:st=stream...](output) — this one creates a program with the title you gave it and the program_num you gave it, then adds the specified stream(s) to it. Cute, tidy, done. ✨
🦋 -stream_group
[map=input_file_id=stream_group][type=type:]st=stream[:st=stream][:stg=stream_group][:id=stream_group_id...](output) — creates a stream group of the type and stream_group_id you asked for, or makes one by mapping an input group, and adds the specified stream(s) and/or previously defined stream_group(s) to it. Very main character energy, honestly.
type can be one of these:
iamf_audio_element — groups the streams that belong to the same IAMF Audio Element.
For this group type, these options are available:
audio_element_type— the Audio Element type. Supported values:channel— scalable channel audio representationscene— ambisonics representation
demixing— demixing information used to reconstruct a scalable channel audio representation. This option must be separated from the rest with a,, and takes thesekey=valueoptions:parameter_id— an identifier that parameters blocks in frames may refer todmixp_mode— a pre-defined combination of demixing parameters
recon_gain— recon gain information used to reconstruct a scalable channel audio representation. This option must be separated from the rest with a,, and takes thesekey=valueoptions:parameter_id— an identifier that parameters blocks in frames may refer to
layer— a layer defining a Channel Layout in the Audio Element. This option must be separated from the rest with a,. Several,-separated entries can be defined, and at least one must be set. It takes these:-separatedkey=valueoptions:ch_layout— the layer's channel layoutflags— the available flags are:recon_gain— whether to signal ifrecon_gainis present as metadata in parameter blocks within frames
output_gainoutput_gain_flags— which channelsoutput_gainapplies to. These flags are available:FL,FR,BL,BR,TFL,TFRambisonics_mode— the ambisonics mode. This has no effect ifaudio_element_typeis set tochannel. Supported values:mono— each ambisonics channel is coded as an individual mono stream in the group
default_w— default weight value
iamf_mix_presentation — groups the streams that belong to all IAMF Audio Elements that the same IAMF Mix Presentation references.
For this group type, these options are available:
submix— a sub-mix within the Mix Presentation. This option must be separated from the rest with a,. Several,-separated entries can be defined, and at least one must be set. It takes these:-separatedkey=valueoptions:parameter_id— an identifier that parameters blocks in frames may refer to, for post-processing the mixed audio signal to generate the audio signal for playbackparameter_rate— the sample rate duration fields in parameters blocks in frames that refer to thisparameter_idare expressed asdefault_mix_gain— default mix gain value to apply when there are no parameter blocks sharing the sameparameter_idfor a given frameelement— references an Audio Element used in this Mix Presentation to generate the final output audio signal for playback. This option must be separated from the rest with a|. Several|-separated entries can be defined, and at least one must be set. It takes these:-separatedkey=valueoptions:stg— thestream_group_idfor an Audio Element which this sub-mix refers toparameter_id— an identifier that parameters blocks in frames may refer to, for applying any processing to the referenced and rendered Audio Element before it gets summed with other processed Audio Elementsparameter_rate— the sample rate duration fields in parameters blocks in frames that refer to thisparameter_idare expressed asdefault_mix_gain— default mix gain value to apply when there are no parameter blocks sharing the sameparameter_idfor a given frameannotations— akey=valuestring describing the sub-mix element, where "key" is a string conforming to BCP-47 that specifies the language for the "value" string. "key" must be the same as the one in the mix's annotationsheadphones_rendering_mode— indicates whether the input channel-based Audio Element is rendered to stereo loudspeakers or spatialized with a binaural renderer when played back on headphones. This has no effect if the referenced Audio Element'saudio_element_typeis set tochannel. Supported values:stereobinaural
layout— specifies the layouts for this sub-mix on which the loudness information was measured. This option must be separated from the rest with a|. Several|-separated entries can be defined, and at least one must be set. It takes these:-separatedkey=valueoptions:layout_typeloudspeakers— the layout follows the loudspeaker sound system convention of ITU-2051-3binaural— the layout is binaural
sound_system— channel layout matching one of Sound Systems A to J of ITU-2051-3, plus7.1.2and3.1.2. This has no effect iflayout_typeis set tobinauralintegrated_loudness— the program integrated loudness information, as defined in ITU-1770-4digital_peak— the digital (sampled) peak value of the audio signal, as defined in ITU-1770-4true_peak— the true peak of the audio signal, as defined in ITU-1770-4dialog_anchored_loudness— the Dialogue loudness information, as defined in ITU-1770-4album_anchored_loudness— the Album loudness information, as defined in ITU-1770-4
annotations— akey=valuestring describing the mix, where "key" is a string conforming to BCP-47 that specifies the language for the "value" string. "key" must be the same as the ones in all sub-mix element's annotations
E.g. to create a scalable 5.1 IAMF file from several WAV input files 🌈
ffmpeg -i front.wav -i back.wav -i center.wav -i lfe.wav
-map 0:0 -map 1:0 -map 2:0 -map 3:0 -c:a opus
-stream_group type=iamf_audio_element:id=1:st=0:st=1:st=2:st=3,
demixing=parameter_id=998,
recon_gain=parameter_id=101,
layer=ch_layout=stereo,
layer=ch_layout=5.1(side),
-stream_group type=iamf_mix_presentation:id=2:stg=0:annotations=en-us=Mix_Presentation,
submix=parameter_id=100:parameter_rate=48000|element=stg=0:parameter_id=100:annotations=en-us=Scalable_Submix|layout=sound_system=stereo|layout=sound_system=5.1(side)
-streamid 0:0 -streamid 1:1 -streamid 2:2 -streamid 3:3 output.iamfTo copy the two stream groups (Audio Element and Mix Presentation) from an input IAMF file with four streams into an mp4 output:
ffmpeg -i input.iamf -c:a copy -stream_group map=0=0:st=0:st=1:st=2:st=3 -stream_group map=0=1:stg=0
-streamid 0:0 -streamid 1:1 -streamid 2:2 -streamid 3:3 output.mp4💜 -target type (output)
Specify the target file type: vcd, svcd, dvd, dv, dv50. type may be prefixed with pal-, ntsc- or film- to use the corresponding standard. All the format options (bitrate, codecs, buffer sizes) then get set automatically for you, which is honestly such a slay. You can just type:
ffmpeg -i myfile.avi -target vcd /tmp/vcd.mpgNevertheless, you can specify additional options as long as you know they do not conflict with the standard, as in:
ffmpeg -i myfile.avi -target vcd -bf 2 /tmp/vcd.mpgso like, the parameters that get set for each -target are literally these, bestie — copying them exactly because no cap, they matter 👇
VCD
<pal>:
-f vcd -muxrate 1411200 -muxpreload 0.44 -packetsize 2324
-s 352x288 -r 25
-codec:v mpeg1video -g 15 -b:v 1150k -maxrate:v 1150k -minrate:v 1150k -bufsize:v 327680
-ar 44100 -ac 2
-codec:a mp2 -b:a 224k<ntsc>:
-f vcd -muxrate 1411200 -muxpreload 0.44 -packetsize 2324
-s 352x240 -r 30000/1001
-codec:v mpeg1video -g 18 -b:v 1150k -maxrate:v 1150k -minrate:v 1150k -bufsize:v 327680
-ar 44100 -ac 2
-codec:a mp2 -b:a 224k<film>:
-f vcd -muxrate 1411200 -muxpreload 0.44 -packetsize 2324
-s 352x240 -r 24000/1001
-codec:v mpeg1video -g 18 -b:v 1150k -maxrate:v 1150k -minrate:v 1150k -bufsize:v 327680
-ar 44100 -ac 2
-codec:a mp2 -b:a 224kSVCD
<pal>:
-f svcd -packetsize 2324
-s 480x576 -pix_fmt yuv420p -r 25
-codec:v mpeg2video -g 15 -b:v 2040k -maxrate:v 2516k -minrate:v 0 -bufsize:v 1835008 -scan_offset 1
-ar 44100
-codec:a mp2 -b:a 224k<ntsc>:
-f svcd -packetsize 2324
-s 480x480 -pix_fmt yuv420p -r 30000/1001
-codec:v mpeg2video -g 18 -b:v 2040k -maxrate:v 2516k -minrate:v 0 -bufsize:v 1835008 -scan_offset 1
-ar 44100
-codec:a mp2 -b:a 224k<film>:
-f svcd -packetsize 2324
-s 480x480 -pix_fmt yuv420p -r 24000/1001
-codec:v mpeg2video -g 18 -b:v 2040k -maxrate:v 2516k -minrate:v 0 -bufsize:v 1835008 -scan_offset 1
-ar 44100
-codec:a mp2 -b:a 224kDVD
<pal>:
-f dvd -muxrate 10080k -packetsize 2048
-s 720x576 -pix_fmt yuv420p -r 25
-codec:v mpeg2video -g 15 -b:v 6000k -maxrate:v 9000k -minrate:v 0 -bufsize:v 1835008
-ar 48000
-codec:a ac3 -b:a 448k<ntsc>:
-f dvd -muxrate 10080k -packetsize 2048
-s 720x480 -pix_fmt yuv420p -r 30000/1001
-codec:v mpeg2video -g 18 -b:v 6000k -maxrate:v 9000k -minrate:v 0 -bufsize:v 1835008
-ar 48000
-codec:a ac3 -b:a 448k<film>:
-f dvd -muxrate 10080k -packetsize 2048
-s 720x480 -pix_fmt yuv420p -r 24000/1001
-codec:v mpeg2video -g 18 -b:v 6000k -maxrate:v 9000k -minrate:v 0 -bufsize:v 1835008
-ar 48000
-codec:a ac3 -b:a 448kDV
<pal>:
-f dv
-s 720x576 -pix_fmt yuv420p -r 25
-ar 48000 -ac 2<ntsc>:
-f dv
-s 720x480 -pix_fmt yuv411p -r 30000/1001
-ar 48000 -ac 2<film>:
-f dv
-s 720x480 -pix_fmt yuv411p -r 24000/1001
-ar 48000 -ac 2okay and the dv50 target is literally identical to the dv target, except the pixel format it sets is yuv422p for all three standards. slay 🫶
💅 lowkey important: any value you set for a parameter up there will override the target preset value. in that case, your output may not comply with the target standard. don't say i didn't warn you.
-dn(input/output) — as an input option, it blocks all data streams of a file from being filtered or being automatically selected or mapped for any output. see the-discardoption if you wanna disable streams individually instead. as an output option, it disables data recording, i.e. automatic selection or mapping of any data stream. want full manual control? see the-mapoption. 💖-dframesnumber (output) — sets the number of data frames to output. it's an obsolete alias for-frames:d, which you should use instead (fr fr, use the new one).-frames[:stream_specifier]framecount (output,per-stream) — stops writing to the stream after framecount frames. bye bestie 👋-q[:stream_specifier]q (output,per-stream)-qscale[:stream_specifier]q (output,per-stream) — use a fixed quality scale (VBR). the meaning of q/qscale is codec-dependent, so like... it depends. if qscale is used without a stream_specifier, then it applies only to the video stream — this is to keep compatibility with previous behaviour, and because specifying the same codec-specific value to 2 different codecs (that's audio and video) generally is not what is intended when no stream_specifier is used. iykyk 🎀-filter[:stream_specifier]filtergraph (output,per-stream) — creates the filtergraph specified by filtergraph and uses it to filter the stream. filtergraph is a description of the filtergraph to apply to the stream, and must have a single input and a single output of the same type of the stream. inside the filtergraph, the input gets associated with the labelin, and the output with the labelout. see theffmpeg-filtersmanual for more tea about the filtergraph syntax ☕ if you want filtergraphs with multiple inputs and/or outputs, check out the-filter_complexoption instead.-reinit_filter[:stream_specifier]integer (input,per-stream) — this boolean option decides if the filtergraph(s) that this stream gets fed into get reinitialized when input frame parameters change mid-stream. it's enabled by default, because most video filters and all audio filters cannot handle deviation in input frame properties. ngl when reinitialization happens, existing filter state is lost — like e.g. the frame countnreference available in some filters. any frames buffered at the time of reinitialization are lost too 😭 the properties where a change triggers reinitialization are: for video, frame resolution or pixel format; for audio, sample format, sample rate, channel count or channel layout.-drop_changed[:stream_specifier]integer (input,per-stream) — this boolean option decides whether a frame with differing frame parameters mid-stream gets dropped instead of leading to filtergraph reinitialization (which would, as established, lead to loss of filter state). generally useful to avoid corrupted yet decodable packets in live streaming inputs. default is false. 🙈-filter_threadsnb_threads (global) — defines how many threads are used to process a filter pipeline. each pipeline will produce a thread pool with this many threads available for parallel processing. the default is the number of available CPUs. multitasking queen 👑-filter_buffered_framesnb_frames (global) — defines the maximum number of buffered frames allowed in a filtergraph. under normal circumstances a filtergraph shouldn't buffer more than a few frames, especially if frames are being fed to it and read from it in a balanced way (which is the intended behavior in ffmpeg). that said, this option lets you limit the total number of frames buffered across all links in a filtergraph. if more frames are generated, filtering is aborted and an error is returned. the default value is0, which means no limit. ✨-pre[:stream_specifier]preset_name (output,per-stream) — specify the preset for matching stream(s). simple as that, bestie.-stats(global) — log encoding progress/statistics asinfo-level log (see-loglevel). it's on by default, so to explicitly disable it you need to specify-nostats. 🎀-stats_periodtime (global) — sets the period at which encoding progress/statistics get updated. default is0.5seconds. tick tock ⏰-print_graphs(global) — prints execution graph details to stderr, in the format set via-print_graphs_format.-print_graphs_filefilename (global) — writes execution graph details to the specified file, in the format set via-print_graphs_format.-print_graphs_formatformat (global) — sets the output format. available formats are:default,compact,csv,flat,ini,json,xml,mermaid,mermaidhtml. the default format isjson. so many choices, main character energy 🌈-progressurl (global) — sends program-friendly progress information to url. progress info is written periodically and at the end of the encoding process. it's made ofkey=valuelines. key consists of only alphanumeric characters. the last key of a sequence of progress information is alwaysprogresswith the valuecontinueorend. the update period is set using-stats_period. for example, log progress information to stdout:
ffmpeg -progress pipe:1 -i in.mkv out.mkv-stdin— enables interaction on standard input. it's on by default unless standard input is used as an input. to explicitly disable interaction you need to specify-nostdin. disabling interaction on standard input is useful, for example, if ffmpeg is in the background process group. roughly the same result can be achieved withffmpeg ... < /dev/nullbut that requires a shell. 💅-debug_ts(global) — prints timestamp/latency information. it's off by default. this option is mostly useful for testing and debugging purposes, and the output format may change from one version to another, so it should not be employed by portable scripts. see also the option-fdebug ts. ⚠️ no cap, don't build scripts on this.-attachfilename (output) — adds an attachment to the output file. this is supported by a few formats like Matroska, for e.g. fonts used in rendering subtitles. attachments are implemented as a specific type of stream, so this option will add a new stream to the file. it's then possible to use per-stream options on this stream in the usual way. attachment streams created with this option will be created after all the other streams (i.e. those created with-mapor automatic mappings). note that for Matroska you also have to set the mimetype metadata tag:
ffmpeg -i INPUT -attach DejaVuSans.ttf -metadata:s:2 mimetype=application/x-truetype-font out.mkv(assuming that the attachment stream will be third in the output file) 🦋
-dump_attachment[:stream_specifier]filename (input,per-stream) — extracts the matching attachment stream into a file named filename. if filename is empty, then the value of thefilenamemetadata tag will be used. e.g. to extract the first attachment to a file namedout.ttf:
ffmpeg -dump_attachment:t:0 out.ttf -i INPUTto extract all attachments to files determined by the filename tag:
ffmpeg -dump_attachment:t "" -i INPUTTechnical note — attachments are implemented as codec extradata, so this option can actually be used to extract extradata from any stream, not just attachments. 💅
🎥 Video Options
Okay bestie, video time. These are the knobs for the moving pictures. 🎬
-vframes number(output) — Set the number of video frames to output. This one is an obsolete alias for-frames:v, and you should use that instead, fr fr.-r[:stream_specifier] fps(input/output, per-stream) — Set frame rate (Hz value, fraction or abbreviation).
As an input option: ignore any timestamps stored in the file and instead generate timestamps assuming constant frame rate fps. This is not the same as the -framerate option used for some input formats like image2 or v4l2 (it used to be the same in older versions of FFmpeg). If in doubt, use -framerate instead of the input option -r.
As an output option:
- video encoding — Duplicate or drop frames right before encoding them to achieve constant output frame rate
fps. - video streamcopy — Indicate to the muxer that
fpsis the stream frame rate. No data is dropped or duplicated in this case. This may produce invalid files iffpsdoes not match the actual stream frame rate as determined by packet timestamps. See also thesettsbitstream filter.
-fpsmax[:stream_specifier] fps(output, per-stream) — Set maximum frame rate (Hz value, fraction or abbreviation).
Clamps output frame rate when output framerate is auto-set and is higher than this value. Useful in batch processing or when input framerate is wrongly detected as very high. It cannot be set together with -r. It is ignored during streamcopy.
-s[:stream_specifier] size(input/output, per-stream) — Set frame size.
As an input option, this is a shortcut for the video_size private option, recognized by some demuxers for which the frame size is either not stored in the file or is configurable — e.g. raw video or video grabbers.
As an output option, this inserts the scale video filter to the end of the corresponding filtergraph. Please use the scale filter directly to insert it at the beginning or some other place.
The format is wxh (default - same as source).
-aspect[:stream_specifier] aspect(output, per-stream) — Set the video display aspect ratio specified byaspect.
aspect can be a floating point number string, or a string of the form num:den, where num and den are the numerator and denominator of the aspect ratio. For example "4:3", "16:9", "1.3333", and "1.7777" are valid argument values.
If used together with -vcodec copy, it will affect the aspect ratio stored at container level, but not the aspect ratio stored in encoded frames, if it exists.
-display_rotation[:stream_specifier] rotation(input, per-stream) — Set video rotation metadata.
rotation is a decimal number specifying the amount in degree by which the video should be rotated counter-clockwise before being displayed.
This option overrides the rotation/display transform metadata stored in the file, if any. When the video is being transcoded (rather than copied) and -autorotate is enabled, the video will be rotated at the filtering stage. Otherwise, the metadata will be written into the output file if the muxer supports it.
If the -display_hflip and/or -display_vflip options are given, they are applied after the rotation specified by this option.
-display_hflip[:stream_specifier](input, per-stream) — Set whether on display the image should be horizontally flipped.
See the -display_rotation option for more details.
-display_vflip[:stream_specifier](input, per-stream) — Set whether on display the image should be vertically flipped.
See the -display_rotation option for more details.
-vn(input/output) — As an input option, blocks all video streams of a file from being filtered or being automatically selected or mapped for any output. See-discardoption to disable streams individually.
As an output option, disables video recording i.e. automatic selection or mapping of any video stream. For full manual control see the -map option.
-vcodec codec(output) — Set the video codec. This is an alias for-codec:v. Slay.-pass[:stream_specifier] n(output, per-stream) — Select the pass number (1or2). It is used to do two-pass video encoding. The statistics of the video are recorded in the first pass into a log file (see also the option-passlogfile), and in the second pass that log file is used to generate the video at the exact requested bitrate. On pass 1, you may just deactivate audio and set output to null, examples for Windows and Unix:
ffmpeg -i foo.mov -c:v libxvid -pass 1 -an -f rawvideo -y NUL
ffmpeg -i foo.mov -c:v libxvid -pass 1 -an -f rawvideo -y /dev/null-passlogfile[:stream_specifier] prefix(output, per-stream) — Set two-pass log file name prefix toprefix, the default file name prefix isffmpeg2pass. The complete file name will bePREFIX-N.log, whereNis a number specific to the output stream.-vf filtergraph(output) — Create the filtergraph specified byfiltergraphand use it to filter the stream.
This is an alias for -filter:v, see the -filter option.
-autorotate— Automatically rotate the video according to file metadata. Enabled by default, use-noautorotateto disable it.-autoscale— Automatically scale the video according to the resolution of first frame. Enabled by default, use-noautoscaleto disable it. When autoscale is disabled, all output frames of filter graph might not be in the same resolution and may be inadequate for some encoder/muxer. Therefore, it is not recommended to disable it unless you really know what you are doing. Disable autoscale at your own risk. 😭
🚀 Advanced Video options
These are the deeper cuts. Still iconic, just a little more niche. ✨
-pix_fmt[:stream_specifier] format(input/output, per-stream) — Set pixel format. Use-pix_fmtsto show all the supported pixel formats. If the selected pixel format can not be selected, ffmpeg will print a warning and select the best pixel format supported by the encoder. Ifpix_fmtis prefixed by a+, ffmpeg will exit with an error if the requested pixel format can not be selected, and automatic conversions inside filtergraphs are disabled. Ifpix_fmtis a single+, ffmpeg selects the same pixel format as the input (or graph output) and automatic conversions are disabled.-sws_flags flags(input/output) — Set default flags for the libswscale library. These flags are used by automatically insertedscalefilters and those within simple filtergraphs, if not overridden within the filtergraph definition.
See the ffmpeg-scaler manual for a list of scaler options.
-rc_override[:stream_specifier] override(output, per-stream) — Rate control override for specific intervals, formatted asint,int,intlist separated with slashes. Two first values are the beginning and end frame numbers, last one is quantizer to use if positive, or quality factor if negative.-vstats— Dump video coding statistics tovstats_HHMMSS.log. See the vstats file format section for the format description.-vstats_file file— Dump video coding statistics tofile. See the vstats file format section for the format description.-vstats_version file— Specify which version of the vstats format to use. Default is2. See the vstats file format section for the format description.-vtag fourcc/tag(output) — Force video tag/fourcc. This is an alias for-tag:v.-force_key_frames[:stream_specifier](output, per-stream) —force_key_framescan take arguments of the following form:time[,time...],expr:expr,source, orscd_metadata.
Full signatures as listed:
-force_key_frames[:stream_specifier] time[,time...]
-force_key_frames[:stream_specifier] expr:expr
-force_key_frames[:stream_specifier] source
-force_key_frames[:stream_specifier] scd_metadatatime[,time...]— If the argument consists of timestamps, ffmpeg will round the specified times to the nearest output timestamp as per the encoder time base and force a keyframe at the first frame having timestamp equal or greater than the computed timestamp. Note that if the encoder time base is too coarse, then the keyframes may be forced on frames with timestamps lower than the specified time. The default encoder time base is the inverse of the output framerate but may be set otherwise via-enc_time_base.
If one of the times is ""chapters"[delta]", it is expanded into the time of the beginning of all chapters in the file, shifted by delta, expressed as a time in seconds. This option can be useful to ensure that a seek point is present at a chapter mark or any other designated place in the output file.
For example, to insert a key frame at 5 minutes, plus key frames 0.1 second before the beginning of every chapter:
-force_key_frames 0:05:00,chapters-0.1expr:expr— If the argument is prefixed withexpr:, the stringexpris interpreted like an expression and is evaluated for each frame. A key frame is forced in case the evaluation is non-zero.
The expression in expr can contain the following constants:
n— the number of current processed frame, starting from 0n_forced— the number of forced framesprev_forced_n— the number of the previous forced frame, it isNANwhen no keyframe was forced yetprev_forced_t— the time of the previous forced frame, it isNANwhen no keyframe was forced yett— the time of the current processed frame
For example to force a key frame every 5 seconds, you can specify:
-force_key_frames expr:gte(t,n_forced*5)To force a key frame 5 seconds after the time of the last forced one, starting from second 13:
-force_key_frames expr:if(isnan(prev_forced_t),gte(t,13),gte(t,prev_forced_t+5))source— If the argument issource, ffmpeg will force a key frame if the current frame being encoded is marked as a key frame in its source. In cases where this particular source frame has to be dropped, enforce the next available frame to become a key frame instead.
scd_metadata— if the argument is literally"scd_metadata", ffmpeg forces a key frame whenever the current frame carries a metadata entry whose key is"lavfi.scd.time". That metadata gets dropped in by filters likescdetandscdet_vulkan💅. Pro tip: don't put frame-duplicating filters afterscdet— you'll end up with duplicate metadata on multiple frames and key frames getting inserted over and over, which is giving chaos honestly 😭.
ngl, forcing way too many keyframes is genuinely harmful to the lookahead algorithms some encoders use — you're better off with fixed-GOP options or something similar, fr fr.
-apply_cropping[:stream_specifier] source (input,per-stream)— auto-crops the video after decoding, following whatever cropping metadata the file says 🌸. Default isall.none(0) — don't apply any cropping metadata at all, leave it be.all(1) — apply both codec- and container-level croppping. This is the default mode, the main character ✨.codec(2) — apply codec level croppping only.container(3) — apply container level croppping only.
-copyinkf[:stream_specifier] (output,per-stream)— when you're stream copying, also copy the non-key frames chillin' at the beginning 🫶.
🧚 hw device stuff (the hardware era)
-init_hw_device type[=name][:device[,key=value...]]— boot up a brand-new hardware device of type type called name, using the device parameters you hand it 🔥. If you don't give a name, it gets a default one shaped liketype%d.
The meaning of device and everything after it depends on the device type, so buckle up bestie:
cuda— device is the number of the CUDA device. Recognised options:primary_ctx— if set to1, uses the primary device context instead of creating a new one.- Examples:
-init_hw_device cuda:1— choose the second device on the system.-init_hw_device cuda:0,primary_ctx=1— choose the first device and use the primary device context.
dxva2— device is the number of the Direct3D 9 display adapter.d3d11va— device is the number of the Direct3D 11 display adapter. If you don't specify it, it attempts to use the default Direct3D 11 display adapter, or the first Direct3D 11 display adapter whose hardware VendorId is specified byvendor_id.- Examples:
-init_hw_device d3d11va— create a d3d11va device on the default Direct3D 11 display adapter.-init_hw_device d3d11va:1— create a d3d11va device on the Direct3D 11 display adapter specified by index 1.-init_hw_device d3d11va:,vendor_id=0x8086— create a d3d11va device on the first Direct3D 11 display adapter whose hardware VendorId is0x8086.
- Examples:
vaapi— device is either an X11 display name, a DRM render node, or a DirectX adapter index 👑. If it's not specified, it attempts to open the default X11 display ($DISPLAY) and then the first DRM render node (/dev/dri/renderD128), or the default DirectX adapter on Windows. Recognised options:kernel_driver— when device isn't specified, use this to name the kernel driver associated with the device you want. Only available when the hardware acceleration methodsdrmandvaapiare enabled.vendor_id— when device andkernel_driveraren't specified, use this to set the vendor id associated with the device you want. Only available when the hardware acceleration methodsdrmandvaapiare enabled andkernel_driverisn't specified.- Examples:
-init_hw_device vaapi— create a vaapi device on the default device.-init_hw_device vaapi:/dev/dri/renderD129— create a vaapi device on DRM render node/dev/dri/renderD129.-init_hw_device vaapi:1— create a vaapi device on DirectX adapter 1.-init_hw_device vaapi:,kernel_driver=i915— create a vaapi device on a device associated with kernel driveri915.-init_hw_device vaapi:,vendor_id=0x8086— create a vaapi device on a device associated with vendor id0x8086.
vdpau— device is an X11 display name. If it's not specified, it attempts to open the default X11 display ($DISPLAY).qsv— device selects a value inMFX_IMPL_*. Allowed values are:auto,sw,hw,auto_any,hw_any,hw2,hw3,hw4. If not specified,auto_anyis used 💀 (tho ngl, it may be easier to get what you want out of QSV by creating the platform-appropriate subdevice —dxva2ord3d11vaorvaapi— and then deriving a QSV device from that). Recognised options:child_device— specify a DRM render node on Linux or a DirectX adapter on Windows.child_device_type— choose a platform-appropriate subdevice type. On Windows,d3d11vais the default subdevice type when--enable-libvplis specified at configuration time, anddxva2is the default subdevice type when--enable-libmfxis specified at configuration time. On Linux you can only usevaapias the subdevice type.- Examples:
-init_hw_device qsv:hw,child_device=/dev/dri/renderD129— create a QSV device withMFX_IMPL_HARDWAREon DRM render node/dev/dri/renderD129.-init_hw_device qsv:hw,child_device=1— create a QSV device withMFX_IMPL_HARDWAREon DirectX adapter 1.-init_hw_device qsv:hw,child_device_type=d3d11va— choose the GPU subdevice with typed3d11vaand create a QSV device withMFX_IMPL_HARDWARE.-init_hw_device qsv:hw,child_device_type=dxva2— choose the GPU subdevice with typedxva2and create a QSV device withMFX_IMPL_HARDWARE.-init_hw_device qsv:hw,child_device=1,child_device_type=d3d11va— create a QSV device withMFX_IMPL_HARDWAREon DirectX adapter 1 with subdevice typed3d11va.-init_hw_device vaapi=va:/dev/dri/renderD129 -init_hw_device qsv=hw1@va— create a VAAPI device calledvaon/dev/dri/renderD129, then derive a QSV device calledhw1from the deviceva. slay 🌈
opencl— device selects the platform and device as platform_index.device_index. You can also filter the set of devices using key-value pairs to find only devices matching particular platform or device strings. The strings usable as filters are:platform_profile,platform_version,platform_name,platform_vendor,platform_extensions,device_name,device_vendor,driver_version,device_version,device_profile,device_extensions,device_type. The indices and filters must together uniquely select a device — no ambiguity allowed bestie ✨.- Examples:
-init_hw_device opencl:0.1— choose the second device on the first platform.-init_hw_device opencl:,device_name=Foo9000— choose the device with a name containing the stringFoo9000.-init_hw_device opencl:1,device_type=gpu,device_extensions=cl_khr_fp16— choose the GPU device on the second platform supporting thecl_khr_fp16extension.
- Examples:
vulkan— if device is an integer, it selects the device by its index in a system-dependent list of devices. If device is any other string, it selects the first device with a name containing that string as a substring 🦋. Recognised options:debug— if set to1, enables the validation layer, if it's installed.linear_images— if set to1, images allocated by the hwcontext will be linear and locally mappable.instance_extensions— a plus-separated list of additional instance extensions to enable.device_extensions— a plus-separated list of additional device extensions to enable.- Examples:
-init_hw_device vulkan:1— choose the second device on the system.-init_hw_device vulkan:RADV— choose the first device with a name containing the stringRADV.-init_hw_device vulkan:0,instance_extensions=VK_KHR_wayland_surface+VK_KHR_xcb_surface— choose the first device and enable the Wayland and XCB instance extensions.
-init_hw_device type[=name]@source— initialise a new hardware device of type type called name, deriving it from the existing device with the name source 🎀.-init_hw_device list— list every hardware device type supported in this build of ffmpeg. Iconic, zero notes.-filter_hw_device name— pass the hardware device called name to all filters in any filter graph. Use it to set the device to upload to with thehwuploadfilter, or the device to map to with thehwmapfilter 💜. Other filters may also use this parameter when they require a hardware device. Note that this is typically only required when the input isn't already in hardware frames — when it is, filters just derive the device they need from the context of the frames they receive as input, iykyk. This is a global setting, so every filter receives the same device.-hwaccel[:stream_specifier] hwaccel (input,per-stream)— use hardware acceleration to decode the matching stream(s) ⚡. The allowed values of hwaccel are:none— don't use any hardware acceleration (the default).auto— automatically select the hardware acceleration method.vdpau— use VDPAU (Video Decode and Presentation API for Unix) hardware acceleration.dxva2— use DXVA2 (DirectX Video Acceleration) hardware acceleration.d3d11va— use D3D11VA (DirectX Video Acceleration) hardware acceleration.vaapi— use VAAPI (Video Acceleration API) hardware acceleration.
qsv— use the Intel QuickSync Video acceleration for video transcoding. okay so heads up: unlike most other values, this one does not enable accelerated decoding (that happens automatically whenever a qsv decoder is selected) — it's accelerated transcoding, without copying the frames into the system memory. For it to work, both the decoder and the encoder must support QSV acceleration and no filters must be used. 💅videotoolbox— use Video Toolbox hardware acceleration.
This option has no effect if the selected hwaccel is not available or not supported by the chosen decoder.
bestie pls read this before you get your hopes up: most acceleration methods are intended for playback and will not be faster than software decoding on modern CPUs 😭. On top of that, ffmpeg will usually need to copy the decoded frames from the GPU memory into the system memory, which is even more performance loss. So like… this option is mainly useful for testing. ngl, that's the whole tea. 🫖
-hwaccel_device[:stream_specifier]hwaccel_device (input,per-stream) — pick a device to use for hardware acceleration. This option only makes sense when the-hwacceloption is also specified ✨. It can either refer to an existing device created with-init_hw_deviceby name, or it can create a new device as if-init_hw_device type:hwaccel_devicewere called immediately before. cute, right? 🎀-hwaccels— list all hardware acceleration components enabled in this build offfmpeg. lowkey important: actual runtime availability depends on the hardware and its suitable driver being installed. so like, being listed ≠ actually working, yk? 💀-fix_sub_duration_heartbeat[:stream_specifier]— set a specific output video stream as the heartbeat stream according to which to split and push through currently in-progress subtitle upon receipt of a random access packet. 🫀
This lowers the latency of subtitles for which the end packet or the following subtitle has not yet been received. As a drawback, this will most likely lead to duplication of subtitle events in order to cover the full duration, so when dealing with use cases where latency of when the subtitle event is passed on to output is not relevant, this option should not be utilized. read that again before you enable it fr fr 🚨
Requires -fix_sub_duration to be set for the relevant input subtitle stream for this to have any effect, as well as for the input subtitle stream having to be directly mapped to the same output in which the heartbeat stream resides. iykyk 💜
🎧 Audio Options 🎧
-aframesnumber (output) — set the number of audio frames to output. This is an obsolete alias for-frames:a, which you should use instead. obsolete means we're not doing that anymore bestie 🙈-ar[:stream_specifier]freq (input/output,per-stream) — set the audio sampling frequency. For output streams it is set by default to the frequency of the corresponding input stream. For input streams this option only makes sense for audio grabbing devices and raw demuxers and is mapped to the corresponding demuxer options. 🔥-aqq (output) — set the audio quality (codec-specific, VBR). This is an alias for-q:a. slay ✨-ac[:stream_specifier]channels (input/output,per-stream) — set the number of audio channels. For output streams it is set by default to the number of input audio channels. For input streams this option only makes sense for audio grabbing devices and raw demuxers and is mapped to the corresponding demuxer options. 🎀-an(input/output) — okay so this one's got two moods 💅:- As an input option, it blocks all audio streams of a file from being filtered or being automatically selected or mapped for any output. See the
-discardoption to disable streams individually. - As an output option, it disables audio recording — i.e. automatic selection or mapping of any audio stream. For full manual control see the
-mapoption. 👑
- As an input option, it blocks all audio streams of a file from being filtered or being automatically selected or mapped for any output. See the
-acodeccodec (input/output) — set the audio codec. This is an alias for-codec:a. ⭐-sample_fmt[:stream_specifier]sample_fmt (output,per-stream) — set the audio sample format. Use-sample_fmtsto get a list of supported sample formats. 🌸-affiltergraph (output) — create the filtergraph specified by filtergraph and use it to filter the stream. This is an alias for-filter:a, see the-filteroption. 🦋
🌟 Advanced Audio options 🌟
-atagfourcc/tag (output) — force audio tag/fourcc. This is an alias for-tag:a. 💖-ch_layout[:stream_specifier]layout (input/output,per-stream) — alias for-channel_layout. 🎀-channel_layout[:stream_specifier]layout (input/output,per-stream) — set the audio channel layout. For output streams it is set by default to the input channel layout. For input streams it overrides the channel layout of the input. noted: not all decoders respect the overridden channel layout 👀. This option also sets the channel layout for audio grabbing devices and raw demuxers and is mapped to the corresponding demuxer option. 🌈-guess_layout_maxchannels (input,per-stream) — if some input channel layout is not known, try to guess only if it corresponds to at most the specified number of channels. For example,2tells toffmpegto recognize 1 channel as mono and 2 channels as stereo but not 6 channels as 5.1. The default is to always try to guess. Use0to disable all guessing. Using the-channel_layoutoption to explicitly specify an input layout also disables guessing. 🔮
📝 Subtitle options 📝
-scodeccodec (input/output) — set the subtitle codec. This is an alias for-codec:s. 💜-sn(input/output) — again, two moods bestie 🫶:- As an input option, it blocks all subtitle streams of a file from being filtered or being automatically selected or mapped for any output. See
-discardoption to disable streams individually. - As an output option, it disables subtitle recording — i.e. automatic selection or mapping of any subtitle stream. For full manual control see the
-mapoption. ✨
- As an input option, it blocks all subtitle streams of a file from being filtered or being automatically selected or mapped for any output. See
🧚 Advanced Subtitle options 🧚
-fix_sub_duration— fix subtitles durations. For each subtitle, wait for the next packet in the same stream and adjust the duration of the first to avoid overlap. This is necessary with some subtitles codecs, especially DVB subtitles, because the duration in the original packet is only a rough estimate and the end is actually marked by an empty subtitle frame. failing to use this option when necessary can result in exaggerated durations or muxing failures due to non-monotonic timestamps 😭💀
⚠️ real talk: this option will delay the output of all data until the next subtitle packet is decoded — it may increase memory consumption and latency a lot. so like, don't just turn it on for fun 🚨
-canvas_sizesize — set the size of the canvas used to render subtitles. 🎨
👑 Advanced options 👑
-map [-]input_file_id[:stream_specifier][:view_specifier][:?] | [linklabel](output) — create one or more streams in the output file. This option has two forms for specifying the data source(s): the first selects one or more streams from some input file (specified with-i), the second takes an output from some complex filtergraph (specified with-filter_complex). 🌟
In the first form, an output stream is created for every stream from the input file with the index input_file_id. If stream_specifier is given, only those streams that match the specifier are used (see the Stream specifiers section for the stream_specifier syntax).
A - character before the stream identifier creates a "negative" mapping. It disables matching streams from already created mappings. 🙅♀️
An optional view_specifier may be given after the stream specifier, which for multiview video specifies the view to be used. The view specifier may have one of the following formats:
view:view_id — select a view by its ID; view_id may be set to'all'to use all the views interleaved into one stream;vidx:view_idx — select a view by its index; i.e.0is the base view,1is the first non-base view, etc.vpos:position — select a view by its display position; position may be"left"or"right"
The default for transcoding is to only use the base view, i.e. the equivalent of vidx:0. For streamcopy, view specifiers are not supported and all views are always copied. 🌀
A trailing ? after the stream index will allow the map to be optional: if the map matches no streams the map will be ignored instead of failing. but wait — note the map will still fail if an invalid input file index is used; such as if the map refers to a non-existent input. 💀
An alternative [linklabel] form will map outputs from complex filter graphs (see the -filter_complex option) to the output file. linklabel must correspond to a defined output link label in the graph. 🔗
This option may be specified multiple times, each adding more streams to the output file. Any given input stream may also be mapped any number of times as a source for different output streams, e.g. in order to use different encoding options and/or filters. The streams are created in the output in the same order in which the -map options are given on the commandline. 👑
Using this option disables the default mappings for this output file. periodt. 💅
Examples:
map everything
To map ALL streams from the first input file to output:
ffmpeg -i INPUT -map 0 outputselect specific stream
If you have two audio streams in the first input file, these streams are identified by 0:0 and 0:1. You can use -map to select which streams to place in an output file. For example:
ffmpeg -i INPUT -map 0:1 out.wavwill map the second input stream in INPUT to the (single) output stream in out.wav. 🎯
create multiple streams
To select the stream with index 2 from input file a.mov (specified by the identifier 0:2), and stream with index 6 from input b.mov (specified by the identifier 1:6), and copy them to the output file out.mov:
ffmpeg -i a.mov -i b.mov -c copy -map 0:2 -map 1:6 out.movcreate multiple streams 2
To select all video and the third audio stream from an input file:
ffmpeg -i INPUT -map 0:v -map 0:a:2 OUTPUTnegative map
To map all the streams except the second audio, use negative mappings:
ffmpeg -i INPUT -map 0 -map -0:a:1 OUTPUToptional map
To map the video and audio streams from the first input, and using the trailing ?, ignore the audio mapping if no audio streams exist in the first input:
ffmpeg -i INPUT -map 0:v -map 0:a? OUTPUT🗣️ map by language 🗣️
So you want the English audio track specifically? Say less, bestie:
ffmpeg -i INPUT -map 0:m:language:eng OUTPUT-ignore_unknown— be chill about input streams whose type nobody can figure out: ignore them instead of failing when someone tries to copy them. Ignorance is bliss 💅.-copy_unknown— opposite energy, and honestly kind of brave: let those unknown-type input streams get copied instead of failing at the attempt. Risk it for the biscuit 🍓.-map_metadata[:metadata_spec_out]infile[:metadata_spec_in](output, per-metadata) — tells the next output file to snag its metadata from infile. Heads up though: those are file indices (zero-based), not filenames, okay? Not "input.mkv", just a number 🔢.
The optional metadata_spec_in/metadata_spec_out bits say which metadata to copy. A metadata specifier can look like this:
g— global metadata, aka the metadata that applies to the whole file.s[:stream_spec]— per-stream metadata. stream_spec is a stream specifier, same deal as described in the Stream specifiers chapter. In an input metadata specifier, the first matching stream gets copied from. In an output metadata specifier, all matching streams get copied to. Yes, that asymmetry is real, ngl.c:chapter_index — per-chapter metadata, where chapter_index is the zero-based chapter index.p:program_index — per-program metadata, where program_index is the zero-based program index.
If you just… don't specify one? It defaults to global. Easy ✨.
By default, global metadata is copied from the first input file, and per-stream and per-chapter metadata gets copied along with its streams/chapters. Those default mappings get switched off the moment you create any mapping of the relevant type. And a negative file index is the pro move for "actually, don't": it creates a dummy mapping that just disables the automatic copying 🙈.
Example time! Copy metadata from the first stream of the input file into the global metadata of the output file:
ffmpeg -i in.ogg -map_metadata 0:s:0 out.mp3And the reverse, bestie — global metadata out to all the audio streams:
ffmpeg -i in.mkv -map_metadata:s:a 0:g out.mkvLowkey a fun fact: a plain 0 would work just as well in that example, since global metadata is assumed by default 👑.
-map_chaptersinput_file_index (output) — copy chapters from the input file with index input_file_index into the next output file. If you specify no chapter mapping at all, chapters get copied from the first input file that has at least one chapter. Want none of that? Negative file index, copy zero chapters, done 🎀.-benchmark(global) — shows benchmarking info at the end of an encode: real, system and user time used, plus maximum memory consumption. Plot twist: max memory consumption isn't supported on all systems, and when it isn't, it'll usually just display as0. Not a bug, it's a vibe check it can't pass 😭.-benchmark_all(global) — same benchmarking energy but during the encode. Shows real, system and user time used in various steps (audio/video encode/decode). For the girlies who need timestamps on their timestamps ⏱️.-timelimitduration (global) — dip out after ffmpeg has been running for duration seconds of CPU user time. Note: user time, not "how long I've been staring at the progress bar" time 💀.-dump(global) — dumps every single input packet tostderr. Every one. Prepare to be humbled by log volume 🌊.-hex(global) — when you're dumping packets, also dump the payload. Hexadecimal on main 🦋.-readratespeed (input) — caps how fast input gets read.
The value is a floating-point positive number, and it represents the maximum duration of media, in seconds, that should be ingested in one second of wallclock time. Default value is zero, which means no imposed limitation on ingestion speed. Value 1 means real-time speed and is equivalent to "-re".
Mainly used to simulate a capture device or a live input stream (like when you're reading from a file). Should not be used with a low value when the input is an actual capture device or live stream — it may cause packet loss. So don't do that, fr fr 🙏.
It's useful when the flow speed of output packets actually matters to you, like in live streaming.
-re(input) — read input at native frame rate. This is equivalent to setting"-readrate 1". Iconic, short, does exactly what it says 🫶.-readrate_initial_burstseconds — set an initial read burst time, in seconds, after which-re/-readrategets enforced. Like a running start before the leash goes on 🏃♀️.-readrate_catchupspeed (input) — if either the input or the output gets blocked and the actual read speed falls behind the specified readrate, then this rate takes effect until the input catches up with the specified readrate. Constraint: it must not be lower than the primary readrate. No sandbagging 🚫.-vsyncparameter (global) — and also-fps_mode[:stream_specifier]parameter (output, per-stream) — sets the video sync method / framerate mode.vsyncapplies to all output video streams but can be overridden for a single stream by settingfps_mode. Also, heads up:vsyncis deprecated and will be removed in the future 💀.
For compatibility reasons, some of the values for vsync can be given as numbers (shown in parentheses below):
passthrough(0) — each frame is passed with its timestamp from the demuxer straight to the muxer. No thoughts, head empty, timestamps intact.cfr(1) — frames get duplicated and dropped to achieve exactly the requested constant frame rate. ✨consistency✨.vfr(2) — frames are passed through with their timestamp, or dropped, so as to prevent 2 frames from having the same timestamp.auto(-1) — chooses betweencfrandvfrdepending on muxer capabilities. This is the default method 👑.
Note that the timestamps may be further modified by the muxer after all this. For example, in the case that the format option avoid_negative_ts is enabled. So like… don't come for me if they shift again.
With -map you can select which stream the timestamps should be taken from. You can leave either video or audio unchanged and sync the remaining stream(s) to the unchanged one. Chef's kiss 🎀.
-frame_drop_thresholdparameter — the frame drop threshold, which specifies how far behind video frames can be before they get dropped. Measured in frame rate units, so1.0is one frame. Default is-1.1. One possible usecase: avoid framedrops in the case of noisy timestamps, or increase frame drop precision in the case of exact timestamps. Choose your fighter 🥊.-apadparameters (output, per-stream) — pad the output audio stream(s). This is the same as applying"-af apad". The argument is a string of filter parameters composed the same way as with the"apad"filter. Big caveat:"-shortest"must be set for this output for the option to take effect. Otherwise it's just sitting there doing nothing, like me at a party 🙈.-copyts— do not process input timestamps; keep their values as-is without trying to sanitize them. In particular, don't remove the initial start time offset value. Feral, but valid.
Note that, depending on the vsync option or on specific muxer processing (e.g. in case the format option avoid_negative_ts is enabled), the output timestamps may mismatch with the input timestamps even when this option is selected. Yeah. Them's the breaks 😭.
-start_at_zero— when used withcopyts, shift input timestamps so they start at zero.
Which means using e.g. "-ss 50" will make output timestamps start at 50 seconds, regardless of what timestamp the input file started at. Clean slate energy ✨.
-copytbmode — specifies how to set the encoder timebase when stream copying. mode is an integer numeric value and can be one of these:1— use the demuxer timebase. The time base is copied to the output encoder from the corresponding input demuxer. This is sometimes required to avoid non monotonically increasing timestamps when copying video streams with variable frame rate. (Say that five times fast.)0— use the decoder timebase. The time base is copied to the output encoder from the corresponding input decoder.-1— try to make the choice automatically, in order to generate a sane output. Let the algorithm cook 🍳.
Default value is -1.
-enc_time_base[:stream_specifier]timebase (output, per-stream) — set the encoder timebase. timebase can be one of these:0— assign a default value according to the media type. For video, use1/framerate; for audio, use1/samplerate.demux— use the timebase from the demuxer.filter— use the timebase from the filtergraph.- a positive number — use the provided number as the timebase. This field can be given as a ratio of two integers (e.g.
1:24,1:48000) or as a decimal number (e.g.0.04166,2.0833e-5).
Default value is 0.
-bitexact(input/output) — enable bitexact mode for (de)muxer and (de/en)coder. Byte-for-byte commitment issues: none 🧚.-shortest(output) — finish encoding when the shortest output stream ends.
Note that this option may require buffering frames, which introduces extra latency. The maximum amount of that latency may be controlled with the "-shortest_buf_duration" option. Good to know before you blame the buffer 💅.
-shortest_buf_durationduration (output) — the"-shortest"option may require buffering potentially large amounts of data when at least one of the streams is "sparse" (i.e. has large gaps between frames — this is typically the case for subtitles, which are drama queens about timing).
This option controls the maximum duration of buffered frames, in seconds. Larger values may allow the "-shortest" option to produce more accurate results, but they increase memory use and latency. Trade-offs! Always trade-offs 😤.
The default value is 10 seconds.
-dts_delta_thresholdthreshold — timestamp discontinuity delta threshold, expressed as a decimal number of seconds.
The timestamp discontinuity correction this enables is only applied to input formats that accept timestamp discontinuity (the ones where the "AVFMT_TS_DISCONT" flag is enabled), e.g. MPEG-TS and HLS. And it's automatically disabled when you're using the "-copyts" option (unless wrapping is detected).
If a timestamp discontinuity is detected whose absolute value is greater than threshold, ffmpeg will remove the discontinuity by decreasing/increasing the current DTS and PTS by the corresponding delta value. Poof, gone ✨.
The default value is 10.
-dts_error_thresholdthreshold — timestamp error delta threshold, expressed as a decimal number of seconds.
This timestamp correction is only applied to input formats that do not accept timestamp discontinuity (the ones where the "AVFMT_TS_DISCONT" flag is not enabled). Different club, different rules.
If a timestamp discontinuity is detected whose absolute value is greater than threshold, ffmpeg will drop the PTS/DTS timestamp value. Yeet 🗑️.
The default value is "3600*30" (30 hours), which is arbitrarily picked and quite conservative. iykyk.
-muxdelayseconds (output) — set the maximum demux-decode delay. That's the whole description, bestie, don't ask me for more 💜.-muxpreloadseconds (output) — sets the initial demux-decode delay. that's it, that's the whole vibe 💅-streamidoutput-stream-index:new-value (output) — slaps a brand new stream-id value onto an output stream. bestie, listen: you gotta put this before the output filename it's meant for, okay? and if you've got multiple output files living their truth, a streamid can absolutely be reassigned to a different value. no cap.
for example, to set the stream 0 PID to 33 and the stream 1 PID to 36 for an output mpegts file:
ffmpeg -i inurl -streamid 0:33 -streamid 1:36 out.ts-bsf[:stream_specifier]bitstream_filters (input/output, per-stream) — applies bitstream filters to the matching streams. the filters hit each packet as it's received from the demuxer (when you use it as an input option) or right before it gets sent to the muxer (when you use it as an output option). ✨
bitstream_filters is a comma-separated list of bitstream filter specs, each one shaped like this:
<filter>[=<optname0>=<optval0>:<optname1>=<optval1>:...]any of the , = : characters that are supposed to be part of an option value need to be escaped with a backslash. yes, really. 🙈
use the -bsfs option to get the list of bitstream filters.
e.g.
ffmpeg -bsf:v h264_mp4toannexb -i h264.mp4 -c:v copy -an out.h264applies the h264_mp4toannexb bitstream filter (it converts an MP4-encapsulated H.264 stream to Annex B) to the input video stream. iconic behaviour honestly.
on the other hand,
ffmpeg -i file.mov -an -vn -bsf:s mov2textsub -c:s copy -f rawvideo sub.txtapplies the mov2textsub bitstream filter (which extracts text from MOV subtitles) to the output subtitle stream. but heads up 🫶 — since both examples use -c copy, it matters very little whether the filters get applied on input or output. that would change if transcoding was happening.
-tag[:stream_specifier]codec_tag (input/output, per-stream) — forces a tag/fourcc for the matching streams. forcing it. with intent. 💅-timecodehh:mm:ssSEPff — specifies the timecode for writing. SEP is:for non drop timecode and;(or.) for drop. lowkey important detail, don't skip it.
ffmpeg -i input.mpg -timecode 01:02:03.04 -r 30000/1001 -s ntsc output.mpg-filter_complexfiltergraph (global) — defines a complex filtergraph, meaning one with an arbitrary number of inputs and/or outputs. for simple graphs — the ones with one input and one output of the same type — go see the-filteroptions instead. filtergraph is a description of the filtergraph, as described in the ``Filtergraph syntax'' section of the ffmpeg-filters manual. this option may be specified multiple times — every single use creates a brand new complex filtergraph. 💖
inputs to a complex filtergraph can come from different source types, and you can tell them apart by the format of the corresponding link label:
- to connect an input stream, use
[file_index:stream_specifier](i.e. the same syntax as-map). if stream_specifier matches multiple streams, the first one will be used. for multiview video, the stream specifier may be followed by the view specifier — see the documentation for the-mapoption for its syntax. - to connect a loopback decoder use
[dec:dec_idx], where dec_idx is the index of the loopback decoder to be connected to the given input. for multiview video, the decoder index may be followed by the view specifier — see the documentation for the-mapoption for its syntax. - to connect an output from another complex filtergraph, use its link label. e.g. the following example:
ffmpeg -i input.mkv \
-filter_complex '[0:v]scale=size=hd1080,split=outputs=2[for_enc][orig_scaled]' \
-c:v libx264 -map '[for_enc]' output.mkv \
-dec 0:0 \
-filter_complex '[dec:0][orig_scaled]hstack[stacked]' \
-map '[stacked]' -c:v ffv1 comparison.mkvreads an input video and
- (line 2) uses a complex filtergraph with one input and two outputs to scale the video to
1920x1080and duplicate the result to both outputs; - (line 3) encodes one scaled output with
libx264and writes the result tooutput.mkv; - (line 4) decodes this encoded stream with a loopback decoder;
- (line 5) places the output of the loopback decoder (i.e. the
libx264-encoded video) side by side with the scaled original input; - (line 6) the combined video is then losslessly encoded and written into
comparison.mkv.
note that the two filtergraphs cannot be combined into one, because then there would be a cycle in the transcoding pipeline (filtergraph output goes to encoding, from there to decoding, then back to the same graph), and such cycles are not allowed. no exceptions bestie, cycles are not the move 💀
an unlabeled input will be connected to the first unused input stream of the matching type.
output link labels are referred to with -map. unlabeled outputs get added to the first output file.
ngl this is the fun part: with this option it's possible to use only lavfi sources without any normal input files. ✨
for example, to overlay an image over video
ffmpeg -i video.mkv -i image.png -filter_complex '[0:v][1:v]overlay[out]' -map '[out]' out.mkvhere [0:v] refers to the first video stream in the first input file, which is linked to the first (main) input of the overlay filter. similarly the first video stream in the second input is linked to the second (overlay) input of overlay.
assuming there is only one video stream in each input file, we can omit input labels, so the above is equivalent to
ffmpeg -i video.mkv -i image.png -filter_complex 'overlay[out]' -map '[out]' out.mkvfurthermore we can omit the output label and the single output from the filter graph will be added to the output file automatically, so we can simply write
ffmpeg -i video.mkv -i image.png -filter_complex 'overlay' out.mkvas a special exception, you can use a bitmap subtitle stream as input: it will be converted into a video with the same size as the largest video in the file, or 720x576 if no video is present. ⚠️ note that this is an experimental and temporary solution — it will be removed once libavfilter has proper support for subtitles. so like, don't build your whole personality around it.
for example, to hardcode subtitles on top of a DVB-T recording stored in MPEG-TS format, delaying the subtitles by 1 second:
ffmpeg -i input.ts -filter_complex \
'[#0x2ef] setpts=PTS+1/TB [sub] ; [#0x2d0] [sub] overlay' \
-sn -map '#0x2dc' output.mkv(0x2d0, 0x2dc and 0x2ef are the MPEG-TS PIDs of respectively the video, audio and subtitles streams; 0:0, 0:3 and 0:7 would have worked too 🫶)
to generate 5 seconds of pure red video using the lavfi color source:
ffmpeg -filter_complex 'color=c=red' -t 5 out.mkv-filter_complex_threadsnb_threads (global) — defines how many threads get used to process afilter_complexgraph. similar tofilter_threadsbut it's used for-filter_complexgraphs only. the default is the number of available CPUs. she's multithreaded and she's thriving 🧚-lavfifiltergraph (global) — define a complex filtergraph, i.e. one with an arbitrary number of inputs and/or outputs. equivalent to-filter_complex. literally the same thing, different outfit 💅-accurate_seek(input) — this option enables or disables accurate seeking in input files with the-ssoption. it's enabled by default, so seeking is accurate when transcoding. use-noaccurate_seekto disable it, which may be useful e.g. when copying some streams and transcoding the others. 🎀-seek_timestamp(input) — this option enables or disables seeking by timestamp in input files with the-ssoption. it's disabled by default. if enabled, the argument to the-ssoption is considered an actual timestamp, and is not offset by the start time of the file. this matters only for files which do not start from timestamp 0, such as transport streams. iykyk.-thread_queue_sizesize (input/output) — for input, this option sets the maximum number of queued packets when reading from the file or device. with low latency / high rate live streams, packets may get discarded if they aren't read in a timely manner; setting this value can force ffmpeg to use a separate input thread and read packets as soon as they arrive. by default ffmpeg only does this if multiple inputs are specified.
for output, this option specifies the maximum number of packets that may be queued to each muxing thread. queue management, but make it cute ✨
-sdp_filefile (global) — print sdp information for an output stream to file. this allows dumping sdp information when at least one output isn't an rtp stream. (requires at least one of the output formats to be rtp, bestie — that part's non-negotiable).-discard(input) — allows discarding specific streams or frames from streams. any input stream can be fully discarded using the valueall, whereas selective discarding of frames from a stream happens at the demuxer and is not supported by all demuxers. 🙈none— discard no frame.default— the default, which discards no frames.noref— discard all non-reference frames.bidir— discard all bidirectional frames.nokey— discard all frames excepts keyframes.all— discard all frames.
-abort_onflags (global) — stop and abort on various conditions. the following flags are available:empty_output— no packets were passed to the muxer, the output is empty.empty_output_stream— no packets were passed to the muxer in some of the output streams.
-max_error_rate(global) — sets the fraction of decoding frame failures across all inputs which, when crossed, makes ffmpeg return exit code69. crossing this threshold does not terminate processing. range is a floating-point number between0and1. default is2/3. nice.-xerror(global) — stop and exit on error. that's it. zero notes. she does not forgive 💀-max_muxing_queue_sizepackets (output, per-stream) — when transcoding audio and/or video streams, ffmpeg will not begin writing into the output until it has one packet for each such stream. while waiting for that to happen, packets for other streams get buffered. this option sets the size of that buffer, in packets, for the matching output stream.
the default value of this option should be high enough for most uses, so only touch this option if you are sure that you need it. hands off the dials unless you know what you're doing, bestie 🫶
-muxing_queue_data_thresholdbytes (output,per-stream) — this is like a minimum threshold, and until you hit it the muxing queue size is just... not taken into account bestie. Defaults to50 megabytesper stream, and it's based on the overall size of packets passed to the muxer.-auto_conversion_filters(global) — turns on automatically inserting format conversion filters in all filter graphs, including the ones defined by-vf,-af,-filter_complexand-lavfi. If filter format negotiation needs a conversion, the initialization of the filters will fail 😭. Conversions can still be performed by inserting the relevant conversion filter (scale,aresample) in the graph. It's on by default, so to explicitly disable it you gotta say-noauto_conversion_filters.-bits_per_raw_sample[:stream_specifier]value (output,per-stream) — declares the number of bits per raw sample in the given output stream to be value. Heads up tho: this option sets the information given to the encoder/muxer, it does not change the stream to conform to this value. Setting values that don't match the stream properties may result in encoding failures or invalid output files 💀.-stats_enc_pre[:stream_specifier]path (output,per-stream) /-stats_enc_post[:stream_specifier]path (output,per-stream) /-stats_mux_pre[:stream_specifier]path (output,per-stream) — write per-frame encoding info about the matching streams into the file given by path.
-stats_enc_pre writes information about raw video or audio frames right before they're sent for encoding, while -stats_enc_post writes information about encoded packets as they're received from the encoder. -stats_mux_pre writes information about packets just as they're about to be sent to the muxer. Every frame or packet produces one line in the specified file. The format of that line is controlled by -stats_enc_pre_fmt / -stats_enc_post_fmt / -stats_mux_pre_fmt.
When stats for multiple streams are written into a single file, the lines corresponding to different streams will be interleaved 🙈. The precise order of this interleaving is not specified and not guaranteed to stay stable between different invocations of the program, even with the same options. So don't come for me if the order shifts, it's giving chaos on purpose.
-stats_enc_pre_fmt[:stream_specifier]format_spec (output,per-stream) /-stats_enc_post_fmt[:stream_specifier]format_spec (output,per-stream) /-stats_mux_pre_fmt[:stream_specifier]format_spec (output,per-stream) — specify the format for the lines written with-stats_enc_pre/-stats_enc_post/-stats_mux_pre.
format_spec is a string that may contain directives of the form {fmt}. format_spec is backslash-escaped — use \{, \}, and \\ to write a literal {, }, or \, respectively, into the output.
The directives given with fmt may be one of these:
fidx— index of the output file.sidx— index of the output stream in the file.n— frame number. Pre-encoding: number of frames sent to the encoder so far. Post-encoding: number of packets received from the encoder so far. Muxing: number of packets submitted to the muxer for this stream so far.ni— input frame number. Index of the input frame (i.e. output by a decoder) that corresponds to this output frame or packet.-1if unavailable.tb— timebase in which this frame/packet's timestamps are expressed, as a rational numbernum/den. Note that encoder and muxer may use different timebases.tbi— timebase forptsi, as a rational numbernum/den. Available whenptsiis available,0/1otherwise.pts— presentation timestamp of the frame or packet, as an integer. Should be multiplied by the timebase to compute presentation time.ptsi— presentation timestamp of the input frame (seeni), as an integer. Should be multiplied bytbito compute presentation time. Printed as (2^63 - 1 = 9223372036854775807) when not available.t— presentation time of the frame or packet, as a decimal number. Equal toptsmultiplied bytb.ti— presentation time of the input frame (seeni), as a decimal number. Equal toptsimultiplied bytbi. Printed asinfwhen not available.dts(packet) — decoding timestamp of the packet, as an integer. Should be multiplied by the timebase to compute presentation time.dt(packet) — decoding time of the frame or packet, as a decimal number. Equal todtsmultiplied bytb.sn(frame,audio) — number of audio samples sent to the encoder so far.samp(frame,audio) — number of audio samples in the frame.size(packet) — size of the encoded packet in bytes.br(packet) — current bitrate in bits per second.abr(packet) — average bitrate for the whole stream so far, in bits per second,-1if it can't be determined at this point.key(packet) — character'K'if the packet contains a keyframe, character'N'otherwise.
Directives tagged with packet may only be used with -stats_enc_post_fmt and -stats_mux_pre_fmt.
Directives tagged with frame may only be used with -stats_enc_pre_fmt.
Directives tagged with audio may only be used with audio streams.
The default format strings are:
- pre-encoding —
{fidx} {sidx} {n} {t} - post-encoding —
{fidx} {sidx} {n} {t}
In the future, new items may be added to the end of the default formatting strings. Users who depend on the format staying exactly the same should prescribe it manually. No cap, write your own format if you're particular about it 💅
Note that stats for different streams written into the same file may have different formats.
🎀 Preset files 🎀
A preset file contains a sequence of option=value pairs, one for each line, specifying a sequence of options which would be awkward to specify on the command line. Lines starting with the hash ('#') character are ignored and are used to provide comments. Check the presets directory in the FFmpeg source tree for examples.
There are two types of preset files: ffpreset and avpreset files.
🌸 ffpreset files 🌸
ffpreset files are specified with the "vpre", "apre", "spre", and "fpre" options. The "fpre" option takes the filename of the preset instead of a preset name as input and can be used for any kind of codec. For the "vpre", "apre", and "spre" options, the options specified in a preset file are applied to the currently selected codec of the same type as the preset option.
The argument passed to the "vpre", "apre", and "spre" preset options identifies the preset file to use according to these rules, bestie, pay attention:
First ffmpeg searches for a file named arg.ffpreset in the directories $FFMPEG_DATADIR (if set), and $HOME/.ffmpeg, and in the datadir defined at configuration time (usually PREFIX/share/ffmpeg) or in a ffpresets folder along the executable on win32, in that order. For example, if the argument is "libvpx-1080p", it will search for the file libvpx-1080p.ffpreset.
If no such file is found, then ffmpeg will search for a file named codec_name-arg.ffpreset in the above-mentioned directories, where codec_name is the name of the codec to which the preset file options will be applied. For example, if you select the video codec with -vcodec libvpx and use -vpre 1080p, then it will search for the file libvpx-1080p.ffpreset. Iconic search behavior honestly ✨
🦋 avpreset files 🦋
avpreset files are specified with the "pre" option. They work similar to ffpreset files, but they only allow encoder-specific options. Therefore, an option=value pair specifying an encoder cannot be used.
When the "pre" option is specified, ffmpeg will look for files with the suffix .avpreset in the directories $AVCONV_DATADIR (if set), and $HOME/.avconv, and in the datadir defined at configuration time (usually PREFIX/share/ffmpeg), in that order.
First ffmpeg searches for a file named codec_name-arg.avpreset in the above-mentioned directories, where codec_name is the name of the codec to which the preset file options will be applied. For example, if you select the video codec with -vcodec libvpx and use -pre 1080p, then it will search for the file libvpx-1080p.avpreset.
If no such file is found, then ffmpeg will search for a file named arg.avpreset in the same directories.
📊 vstats file format 📊
The -vstats and -vstats_file options enable generation of a file containing statistics about the generated video outputs.
The -vstats_version option controls the format version of the generated file.
With version 1 the format is:
frame= <FRAME> q= <FRAME_QUALITY> PSNR= <PSNR> f_size= <FRAME_SIZE> s_size= <STREAM_SIZE>kB time= <TIMESTAMP> br= <BITRATE>kbits/s avg_br= <AVERAGE_BITRATE>kbits/sWith version 2 the format is:
out= <OUT_FILE_INDEX> st= <OUT_FILE_STREAM_INDEX> frame= <FRAME_NUMBER> q= <FRAME_QUALITY>f PSNR= <PSNR> f_size= <FRAME_SIZE> s_size= <STREAM_SIZE>kB time= <TIMESTAMP> br= <BITRATE>kbits/s avg_br= <AVERAGE_BITRATE>kbits/sThe value corresponding to each key is described below:
avg_br— average bitrate expressed in Kbits/sbr— bitrate expressed in Kbits/sframe— number of encoded frameout— out file indexPSNR— Peak Signal to Noise Ratioq— quality of the framef_size— encoded packet size expressed as number of bytess_size— stream size expressed in KiBst— out file stream indextime— time of the packettype— picture type
See also the -stats_enc options for an alternative way to show encoding statistics 🔥
✨ EXAMPLES ✨
okay bestie this is the part where it all comes together — real commands, real vibes, copy-paste ready 🫶
🎥 Video and Audio grabbing
so if you specify the input format and the device, ffmpeg can grab video and audio directly. no middleman, no drama:
ffmpeg -f oss -i /dev/dsp -f video4linux2 -i /dev/video0 /tmp/out.mpgor, if you'd rather use an ALSA audio source (mono input, card id 1) instead of OSS:
ffmpeg -f alsa -ac 1 -i hw:1 -f video4linux2 -i /dev/video0 /tmp/out.mpgheads up 💅 — you must activate the right video source and channel before launching ffmpeg with any TV viewer such as <http://linux.bytesex.org/xawtv/> by Gerd Knorr. like, the viewer's gotta be doing its thing first. you also have to set the audio recording levels correctly with a standard mixer, no cap.
🖥️ X11 grabbing
grab the X11 display with ffmpeg via:
ffmpeg -f x11grab -video_size cif -framerate 25 -i :0.0 /tmp/out.mpg0.0 is the display.screen number of your X11 server, same as the DISPLAY environment variable. iconic.
ffmpeg -f x11grab -video_size cif -framerate 25 -i :0.0+10,20 /tmp/out.mpgagain, 0.0 is the display.screen number of your X11 server, same as the DISPLAY environment variable. 10 is the x-offset and 20 the y-offset for the grabbing. offsets slay 💅
🔄 Video and Audio file format conversion
any supported file format and protocol can serve as input to ffmpeg. literally any.
Examples:
- you can use YUV files as input:
ffmpeg -i /tmp/test%d.Y /tmp/out.mpgit will use the files:
/tmp/test0.Y, /tmp/test0.U, /tmp/test0.V,
/tmp/test1.Y, /tmp/test1.U, /tmp/test1.V, etc...the Y files use twice the resolution of the U and V files. they're raw files, without header. they can be generated by all decent video decoders. you must specify the size of the image with the -s option if ffmpeg cannot guess it 👑
- you can input from a raw YUV420P file:
ffmpeg -i /tmp/test.yuv /tmp/out.avitest.yuv is a file containing raw YUV planar data. each frame is composed of the Y plane followed by the U and V planes at half vertical and horizontal resolution. planar, babes 💜
- you can output to a raw YUV420P file:
ffmpeg -i mydivx.avi hugefile.yuv- you can set several input files and output files:
ffmpeg -i /tmp/a.wav -s 640x480 -i /tmp/a.yuv /tmp/a.mpgconverts the audio file a.wav and the raw YUV video file a.yuv to MPEG file a.mpg. multi-tasking queen ✨
- you can also do audio and video conversions at the same time:
ffmpeg -i /tmp/a.wav -ar 22050 /tmp/a.mp2converts a.wav to MPEG audio at 22050 Hz sample rate.
- you can encode to several formats at the same time and define a mapping from input stream to output streams:
ffmpeg -i /tmp/a.wav -map 0:a -b:a 64k /tmp/a.mp2 -map 0:a -b:a 128k /tmp/b.mp2converts a.wav to a.mp2 at 64 kbits and to b.mp2 at 128 kbits. -map file:index specifies which input stream is used for each output stream, in the order of the definition of output streams. yeah, the ordering matters, iykyk.
- you can transcode decrypted VOBs:
ffmpeg -i snatch_1.vob -f avi -c:v mpeg4 -b:v 800k -g 300 -bf 2 -c:a libmp3lame -b:a 128k snatch.avithis is a typical DVD ripping example — the input is a VOB file, the output an AVI file with MPEG-4 video and MP3 audio. note that in this command we use B-frames so the MPEG-4 stream is DivX5 compatible, and GOP size is 300 which means one intra frame every 10 seconds for 29.97fps input video. furthermore, the audio stream is MP3-encoded so you need to enable LAME support by passing --enable-libmp3lame to configure. the mapping is particularly useful for DVD transcoding to get the desired audio language 🌈
NOTE: to see the supported input formats, use ffmpeg -demuxers. 🎀
- you can extract images from a video, or create a video from many images:
for extracting images from a video:
ffmpeg -i foo.avi -r 1 -s WxH -f image2 foo-%03d.jpegthis will extract one video frame per second from the video and will output them in files named foo-001.jpeg, foo-002.jpeg, etc. images will be rescaled to fit the new WxH values.
if you want to extract just a limited number of frames, you can use the above command in combination with the -frames:v or -t option, or in combination with -ss to start extracting from a certain point in time. lowkey chef's kiss 👑
for creating a video from many images:
ffmpeg -f image2 -framerate 12 -i foo-%03d.jpeg -s WxH foo.avithe syntax foo-%03d.jpeg specifies to use a decimal number composed of three digits padded with zeroes to express the sequence number. it is the same syntax supported by the C printf function, but only formats accepting a normal integer are suitable. fr fr, don't get cute with it.
when importing an image sequence, -i also supports expanding shell-like wildcard patterns (globbing) internally, by selecting the image2-specific -pattern_type glob option.
for example, for creating a video from filenames matching the glob pattern foo-*.jpeg:
ffmpeg -f image2 -pattern_type glob -framerate 12 -i 'foo-*.jpeg' -s WxH foo.avi- you can put many streams of the same type in the output:
ffmpeg -i test1.avi -i test2.avi -map 1:1 -map 1:0 -map 0:1 -map 0:0 -c copy -y test12.nutthe resulting output file test12.nut will contain the first four streams from the input files in reverse order. chaotic but correct 💀
- to force CBR video output:
ffmpeg -i myfile.avi -b 4000k -minrate 4000k -maxrate 4000k -bufsize 1835k out.m2v- the four options
lmin,lmax,mblminandmblmaxuse 'lambda' units, but you may use theQP2LAMBDAconstant to easily convert from 'q' units:
ffmpeg -i src.ext -lmax 21*QP2LAMBDA dst.ext👀 SEE ALSO 👀
ffmpeg-all(1), ffplay(1), ffprobe(1), ffmpeg-utils(1), ffmpeg-scaler(1), ffmpeg-resampler(1), ffmpeg-codecs(1), ffmpeg-bitstream-filters(1), ffmpeg-formats(1), ffmpeg-devices(1), ffmpeg-protocols(1), ffmpeg-filters(1)
like, the whole friend group 🫶 go read them too.
💖 AUTHORS 💖
The FFmpeg developers. the main characters, honestly.
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. very organized of them ✨
and that's a wrap on the whole page, bestie 🎀 this was FFMPEG(1) from the FFmpeg docs — the version and date live in that little footer stamped at the bottom of the real page, so go peep your installed copy for the exact source/version and date 💜 now go forth and transcode something iconic. periodt. 💅🔥
📦 where this page comes from
- alpine 3.24: ffmpeg-doc 8.1.3-r0 · GPL-2.0-or-later AND LGPL-2.1-or-later
- nixos 26.05: ffmpeg 8.1.3
- nixos 26.05: ffmpeg-full 8.1.3
- nixos 26.05: ffmpeg-headless 8.1.3
- nixpkgs unstable: jellyfin-ffmpeg 8.1.3-1
- 💖 girlified by deepseek-ai/DeepSeek-V4.1-Flash-1017 on : a language model's rewrite of the original, under its terms