FFmpeg is a powerful tool for manipulating audiovisual files. Unfortunately, it also has a steep learning curve, especially for users unfamiliar with a command line interface. This app helps users through the command generation process so that more people can reap the benefits of FFmpeg.
Each button displays helpful information about how to perform a wide variety of tasks using FFmpeg. To use this site, click on the task you would like to perform. A new window will open up with a sample command and a description of how that command works. You can copy this command and understand how the command works with a breakdown of each of the flags.
This page does not have search functionality, but you can open all recipes (second option in the sidebar) and use your browser's search tool (often ctrl+f or cmd+f) to perform a keyword search through all recipes.
For FFmpeg basics, check out the program’s official website.
For instructions on how to install FFmpeg on Mac, Linux, and Windows, refer to Reto Kromer’s installation instructions.
For Bash and command line basics, try the Command Line Crash Course. For a little more context presented in an ffmprovisr style, try explainshell.com!
This work is licensed under a Creative Commons Attribution 4.0 International License.
Script Ahoy: Community Resource for Archivists and Librarians Scripting
The Sourcecaster: an app that helps you use the command line to work through common challenges that come up when working with digital primary sources.
Cable Bible: A Guide to Cables and Connectors Used for Audiovisual Tech
At its basis, an FFmpeg command is relatively simple. After you have installed FFmpeg (see instructions here), the program is invoked simply by typing ffmpeg at the command prompt.
ffmpeg
Subsequently, each instruction that you supply to FFmpeg is actually a pair: a flag, which designates the type of action you want to carry out; and then the specifics of that action. Flags are always prepended with a hyphen.
For example, in the instruction -i input_file.ext, the -i flag tells FFmpeg that you are supplying an input file, and input_file.ext states which file it is.
-i input_file.ext
-i
input_file.ext
Likewise, in the instruction -c:v prores, the flag -c:v tells FFmpeg that you want to encode the video stream, and prores specifies which codec is to be used. (-c:v is shorthand for -codec:v/-codec:video).
-c:v prores
-c:v
prores
-codec:v
-codec:video
A very basic FFmpeg command looks like this:
FFplay allows you to stream created video and FFmpeg allows you to save video.
The following command creates and saves a 10-second video of SMPTE bars:
ffmpeg -f lavfi -i smptebars=size=640x480 -t 5 output_file
This command plays and streams SMPTE bars but does not save them on the computer:
ffplay -f lavfi smptebars=size=640x480
The main difference is small but significant: the -i flag is required for FFmpeg but not required for FFplay. Additionally, the FFmpeg script needs to have -t 5 and output.mkv added to specify the length of time to record and the place to save the video.
-t 5
output.mkv
Unless specified, FFmpeg will automatically set codec choices and codec parameters based off of internal defaults. These defaults are applied based on the file type used in the output (for example .mov or .wav).
.mov
.wav
When creating or transcoding files with FFmpeg, it is important to consider codec settings for both audio and video, as the default options may not be desirable in your particular context. The following is a brief list of codec defaults for some common file types:
.avi
.mkv
.mp4
.mpg
Many FFmpeg commands use filters that manipulate the video or audio stream in some way: for example, hflip to horizontally flip a video, or amerge to merge two or more audio tracks into a single stream.
The use of a filter is signalled by the flag -vf (video filter) or -af (audio filter), followed by the name and options of the filter itself. For example, take the convert colourspace command:
-vf
-af
ffmpeg -i input_file -c:v libx264 -vf colormatrix=src:dst output_file
Here, colormatrix is the filter used, with src and dst representing the source and destination colourspaces. This part following the -vf is a filtergraph.
It is also possible to apply multiple filters to an input, which are sequenced together in the filtergraph. A chained set of filters is called a filter chain, and a filtergraph may include multiple filter chains. Filters in a filterchain are separated from each other by commas (,), and filterchains are separated from each other by semicolons (;). For example, take the inverse telecine command:
,
;
ffmpeg -i input_file -c:v libx264 -vf "fieldmatch,yadif,decimate" output_file
Here we have a filtergraph including one filter chain, which is made up of three video filters.
It is often prudent to enclose your filtergraph in quotation marks; this means that you can use spaces within the filtergraph. Using the inverse telecine example again, the following filter commands are all valid and equivalent:
-vf fieldmatch,yadif,decimate
-vf "fieldmatch,yadif,decimate"
-vf "fieldmatch, yadif, decimate"
but -vf fieldmatch, yadif, decimate is not valid.
-vf fieldmatch, yadif, decimate
The ordering of the filters is significant. Video filters are applied in the order given, with the output of one filter being passed along as the input to the next filter in the chain. In the example above, fieldmatch reconstructs the original frames from the inverse telecined video, yadif deinterlaces (this is a failsafe in case any combed frames remain, for example if the source mixes telecined and real interlaced content), and decimate deletes duplicated frames. Clearly, it is not possible to delete duplicated frames before those frames are reconstructed.
fieldmatch
yadif
decimate
-filter:v
-filter_complex
For more information, check out the FFmpeg wiki Filtering Guide.
Stream mapping is the practice of defining which of the streams (e.g., video or audio tracks) present in an input file will be present in the output file. FFmpeg recognises five stream types:
a
v
s
d
t
Mapping is achieved by use of the -map flag, followed by an action of the type file_number:stream_type[:stream_number]. Numbering is zero-indexed, and it's possible to map by stream type and/or overall stream order within the input file. For example:
-map
file_number:stream_type[:stream_number]
-map 0:v
-map 0:3
-map 0:a:2
-map 0:0 -map 0:2
-map 0:1 -map 1:0
To map all streams in the input file to the output file, use -map 0. However, note that not all container formats can include all stream types: for example, .mp4 cannot contain timecode.
-map 0
When no mapping is specified in an ffmpeg command, the default for video files is to take just one video and one audio stream for the output: other stream types, such as timecode or subtitles, will not be copied to the output file by default. If multiple video or audio streams are present, the best quality one is automatically selected by FFmpeg.
For more information, check out the FFmpeg wiki Map page, and the official FFmpeg documentation on -map.
ffmpeg -i input_file.ext -c copy -map 0 output_file.ext
This script will rewrap a video file. It will create a new video video file where the inner content (the video, audio, and subtitle data) of the original file is unchanged, but these streams are rehoused within a different container format.
Note: rewrapping is also known as remuxing, short for re-multiplexing.
It may not be possible to rewrap a file's contents to a new container without re-encoding one or more of the streams within (that is, the video, audio, and subtitle tracks). Some containers can only contain streams of a certain encoding type: for example, the .mp4 container does not support uncompressed audio tracks. (In practice .mp4 goes hand-in-hand with a H.264-encoded video stream and an AAC-encoded video stream, although other types of video and audio streams are possible). Another example is that the Matroska container does not allow data tracks; see the MKV to MP4 recipe.
In such cases, FFmpeg will throw an error. If you encounter errors of this kind, you may wish to consult the list of transcoding recipes.
ffmpeg -i input_file.mkv -c:v copy -c:a aac output_file.mp4
This will convert your Matroska (MKV) files to MP4 files.
-c:a aac
-an
ffmpeg -i input_file -f rawvideo -c:v copy output_file.dv
This script will take a video that is encoded in the DV Codec but wrapped in a different container (such as MOV) and rewrap it into a raw DV file (with the .dv extension). Since DV files potentially contain a great deal of provenance metadata within the DV stream, it is necessary to rewrap files in this method to avoid unintentional stripping of this metadata.
ffmpeg -i input_file -c:v prores -profile:v 1 -vf yadif -c:a pcm_s16le output_file.mov
This command transcodes an input file into a deinterlaced Apple ProRes 422 LT file with 16-bit linear PCM encoded audio. The file is deinterlaced using the yadif filter (Yet Another De-Interlacing Filter).
FFmpeg comes with more than one ProRes encoder:
prores_ks
-c:v prores_ks -profile:v 4
-c:v prores_ks -profile:v 5
ffmpeg -i input_file -c:v libx264 -pix_fmt yuv420p -c:a aac output_file
This command takes an input file and transcodes it to H.264 with an .mp4 wrapper, audio is transcoded to AAC. The libx264 codec defaults to a “medium” preset for compression quality and a CRF of 23. CRF stands for constant rate factor and determines the quality and file size of the resulting H.264 video. A low CRF means high quality and large file size; a high CRF means the opposite.
In order to optimize the file for streaming, you can add this preset:
ffmpeg -i input_file -c:v libx264 -pix_fmt yuv420p -c:a aac -movflags +faststart output_file
In order to use the same basic command to make a higher quality file, you can add some of these presets:
ffmpeg -i input_file -c:v libx264 -pix_fmt yuv420p -preset veryslow -crf 18 -c:a aac output_file
veryslow
slower
slow
medium
fast
faster
veryfast
superfast
ultrafast
-pix_fmt yuv420p
libx264
For more information, see the FFmpeg and H.264 Encoding Guide on the FFmpeg wiki.
ffmpeg -i input_video_file.mxf -i input_audio_file.mxf -c:v libx264 -pix_fmt yuv420p -c:a aac output_file.mp4
This will transcode MXF wrapped video and audio files to an H.264 encoded MP4 file. Please note this only works for unencrypted, single reel DCPs.
.mxf
Variation: Copy PCM audio streams by using Matroska instead of the MP4 container
ffmpeg -i input_video_file.mxf -i input_audio_file.mxf -c:v libx264 -pix_fmt yuv420p -c:a copy output_file.mkv
ffmpeg -i input_file -map 0 -dn -c:v ffv1 -level 3 -g 1 -slicecrc 1 -slices 16 -c:a copy output_file.mkv -f framemd5 -an framemd5_output_file
This will losslessly transcode your video with the FFV1 Version 3 codec in a Matroska container. In order to verify losslessness, a framemd5 of the source video is also generated. For more information on FFV1 encoding, try the FFmpeg wiki.
ffmpeg -i concat:input_file_1\|input_file_2\|input_file_3 -c:v libx264 -c:a aac output_file.mp4
This command allows you to create an H.264 file from a DVD source that is not copy-protected.
Before encoding, you’ll need to establish which of the .VOB files on the DVD or .iso contain the content that you wish to encode. Inside the VIDEO_TS directory, you will see a series of files with names like VTS_01_0.VOB, VTS_01_1.VOB, etc. Some of the .VOB files will contain menus, special features, etc, so locate the ones that contain target content by playing them back in VLC.
-i concat:VTS_01_1.VOB\|VTS_01_2.VOB\|VTS_01_3.VOB
It’s also possible to adjust the quality of your output by setting the -crf and -preset values:
ffmpeg -i concat:input_file_1\|input_file_2\|input_file_3 -c:v libx264 -crf 18 -preset veryslow -c:a aac output_file.mp4
Bear in mind that by default, libx264 will only encode a single video stream and a single audio stream, picking the ‘best’ of the options available. To preserve all video and audio streams, add -map parameters:
ffmpeg -i concat:input_file_1\|input_file_2 -map 0:v -map 0:a -c:v libx264 -c:a aac output_file.mp4
ffmpeg -i input_file -c:v libx265 -pix_fmt yuv420p -c:a copy output_file
This command takes an input file and transcodes it to H.265/HEVC in an .mp4 wrapper, keeping the audio codec the same as in the original file.
Note: FFmpeg must be compiled with libx265, the library of the H.265 codec, for this script to work. (Add the flag --with-x265 if using the brew install ffmpeg method).
--with-x265
brew install ffmpeg
The libx265 encoding library defaults to a ‘medium’ preset for compression quality and a CRF of 28. CRF stands for ‘constant rate factor’ and determines the quality and file size of the resulting H.265 video. The CRF scale ranges from 0 (best quality [lossless]; largest file size) to 51 (worst quality; smallest file size).
A CRF of 28 for H.265 can be considered a medium setting, corresponding to a CRF of 23 in encoding H.264, but should result in about half the file size.
To create a higher quality file, you can add these presets:
ffmpeg -i input_file -c:v libx265 -pix_fmt yuv420p -preset veryslow -crf 18 -c:a copy output_file
ffmpeg -i input_file -acodec libvorbis -b:v 690k output_file
This command takes an input file and transcodes it to Ogg/Theora in an .ogv wrapper with 690k video bitrate.
Note: FFmpeg must be installed with support for Ogg Theora. If you are using Homebrew, you can check with brew info ffmpeg and then update it with brew upgrade ffmpeg --with-theora --with-libvorbis if necessary.
brew info ffmpeg
brew upgrade ffmpeg --with-theora --with-libvorbis
.ogv
This recipe is based on Paul Rouget's recipes.
ffmpeg -i input_file.wav -write_id3v1 1 -id3v2_version 3 -dither_method rectangular -out_sample_rate 48k -qscale:a 1 output_file.mp3
This will convert your WAV files to MP3s.
-b:a 320k
A couple notes
ffmpeg -i input_file -i input_file_to_append -filter_complex "[0:a:0]asplit=2[a][b];[b]afifo[bb];[1:a:0][bb]concat=n=2:v=0:a=1[concatout]" -map "[a]" -codec:a libmp3lame -dither_method modified_e_weighted -qscale:a 2 output_file.mp3 -map "[concatout]" -codec:a libmp3lame -dither_method modified_e_weighted -qscale:a 2 output_file_appended.mp3
This script allows you to generate two derivative audio files from a master while appending audio from a separate file (for example a copyright or institutional notice) to one of them.
asplit
concat
n=2
v=0:a=1
ffmpeg -i input_file.wav -c:a aac -b:a 128k -dither_method rectangular -ar 44100 output_file.mp4
This will convert your WAV file to AAC/MP4.
A note about dither methods. FFmpeg comes with a variety of dither algorithms, outlined in the official docs, though some may lead to unintended, not-subtle digital clipping on some systems.
Transform a video file with 4:3 aspect ratio into a video file with 16:9 aspect ratio by correct pillarboxing.
ffmpeg -i input_file -filter:v "pad=ih*16/9:ih:(ow-iw)/2:(oh-ih)/2" -c:a copy output_file
-c:a copy
Transform a video file with 16:9 aspect ratio into a video file with 4:3 aspect ratio by correct letterboxing.
ffmpeg -i input_file -filter:v "pad=iw:iw*3/4:(ow-iw)/2:(oh-ih)/2" -c:a copy output_file
ffmpeg -i input_file -filter:v "hflip,vflip" -c:a copy output_file
Transform a SD video file with 4:3 aspect ratio into an HD video file with 16:9 aspect ratio by correct pillarboxing.
ffmpeg -i input_file -filter:v "colormatrix=bt601:bt709, scale=1440:1080:flags=lanczos, pad=1920:1080:240:0" -c:a copy output_file
scale=1440:1080
flags=lanczos
pad=1920:1080:240:0
ffmpeg -i input_file -c:v copy -aspect 4:3 output_file
4:3
16:9
-c:v copy
This command uses a filter to convert the video to a different colour space.
bt601
smpte170m
bt470bg
bt709
bt2020
-vf colormatrix=bt601:bt709
Note: Converting between colourspaces with FFmpeg can be done via either the colormatrix or colorspace filters, with colorspace allowing finer control (individual setting of colourspace, transfer characteristics, primaries, range, pixel format, etc). See this entry on the FFmpeg wiki, and the FFmpeg documentation for colormatrix and colorspace.
ffmpeg -i input_file -c:v libx264 -vf colormatrix=src:dst -color_primaries val -color_trc val -colorspace val output_file
gamma28
bt2020_10
bt2020_12
bt2020_cl
bt2020_ncl
To Rec.601 (525-line/NTSC):
ffmpeg -i input_file -c:v libx264 -vf colormatrix=bt709:smpte170m -color_primaries smpte170m -color_trc smpte170m -colorspace smpte170m output_file
To Rec.601 (625-line/PAL):
ffmpeg -i input_file -c:v libx264 -vf colormatrix=bt709:bt470bg -color_primaries bt470bg -color_trc gamma28 -colorspace bt470bg output_file
To Rec.709:
ffmpeg -i input_file -c:v libx264 -vf colormatrix=bt601:bt709 -color_primaries bt709 -color_trc bt709 -colorspace bt709 output_file
MediaInfo output examples:
⚠ Using this command it is possible to add Rec.709 tags to a file that is actually Rec.601 (etc), so apply with caution!
These commands are relevant for H.264 and H.265 videos, encoded with libx264 and libx265 respectively.
libx265
Note: If you wish to embed colourspace metadata without changing to another colourspace, omit -vf colormatrix=src:dst. However, since it is libx264/libx265 that writes the metadata, it’s not possible to add these tags without reencoding the video stream.
-vf colormatrix=src:dst
For all possible values for -color_primaries, -color_trc, and -colorspace, see the FFmpeg documentation on codec options.
-color_primaries
-color_trc
-colorspace
1. Out of step with the regular pattern, -color_trc doesn’t accept bt470bg; it is instead here referred to directly as gamma. In the Rec.601 standard, 525-line/NTSC and 625-line/PAL video have assumed gammas of 2.2 and 2.8 respectively. ↩
E.g. for converting 24fps to 25fps with audio pitch compensation for PAL access copies. (Thanks @kieranjol!)
ffmpeg -i input_file -r output_fps -filter_complex "[0:v]setpts=input_fps/output_fps*PTS[v]; [0:a]atempo=output_fps/input_fps[a]" -map "[v]" -map "[a]" output_file
setpts
atempo
input_fps
output_fps
These examples use QuickTime inputs and outputs. The strategy will vary or may not be possible in other file formats. In the case of these examples it is the intention to make a lossless copy while clarifying an unknown characteristic of the stream.
ffprobe input_file -show_streams
Values that are set to 'unknown' and 'undetermined' may be unspecified within the stream. An unknown aspect ratio would be expressed as '0:1'. Streams with many unknown properties may have interoperability issues or not play as intended. In many cases, an unknown or undetermined value may be accurate because the information about the source is unclear, but often the value is intended to be known. In many cases the stream will played with an assumed value if undetermined (for instance a display_aspect_ratio of '0:1' may be played as 'WIDTH:HEIGHT'), but this may or may not be what is intended. Use carefully.
If the display_aspect_ratio is set to '0:1' it may be clarified with the -aspect option and stream copy.
ffmpeg -i input_file -c copy -map 0 -aspect DAR_NUM:DAR_DEN output_file
Other properties may be clarified in a similar way. Replace -aspect and its value with other properties such as shown in the options below. Note that setting color values in QuickTime requires that -movflags write_colr is set.
The possible values for -color_primaries, -color_trc, and -field_order are given in the Codec Options section of the FFmpeg docs - scroll down to near the bottom of the section.
-field_order
ffmpeg -i input_file -vf "crop=width:height" output_file
This command crops the input video to the dimensions defined
It's also possible to specify the crop position by adding the x and y coordinates representing the top left of your cropped area to your crop filter, as such:
ffmpeg -i input_file -vf "crop=width:height[:x_position:y_position]" output_file
The original frame, a screenshot of the SMPTE colourbars:
Result of the command ffmpeg -i smpte_coloursbars.mov -vf "crop=500:500" output_file:
ffmpeg -i smpte_coloursbars.mov -vf "crop=500:500" output_file
Result of the command ffmpeg -i smpte_coloursbars.mov -vf "crop=500:500:0:0" output_file, appending :0:0 to crop from the top left corner:
ffmpeg -i smpte_coloursbars.mov -vf "crop=500:500:0:0" output_file
:0:0
Result of the command ffmpeg -i smpte_coloursbars.mov -vf "crop=500:300:500:30" output_file:
ffmpeg -i smpte_coloursbars.mov -vf "crop=500:300:500:30" output_file
ffmpeg -i input_file -c:a copy -vn output_file
This command extracts the audio stream without loss from an audiovisual file.
ffmpeg -i input_file -filter_complex "[0:a:0][0:a:1]amerge[out]" -map 0:v -map "[out]" -c:v copy -shortest output_file
This command combines two audio tracks present in a video file into one stream. It can be useful in situations where a downstream process, like YouTube’s automatic captioning, expect one audio track. To ensure that you’re mapping the right audio tracks run ffprobe before writing the script to identify which tracks are desired. More than two audio streams can be combined by extending the pattern present in the -filter_complex option.
ffmpeg -i input_file -af pan="stereo|c0=c0|c1=-1*c1" output_file
This command inverses the audio phase of the second channel by rotating it 180°.
ffmpeg -i input_file -af loudnorm=print_format=json -f null -
This filter calculates and outputs loudness information in json about an input file (labeled input) as well as what the levels would be if loudnorm were applied in its one pass mode (labeled output). The values generated can be used as inputs for a 'second pass' of the loudnorm filter allowing more accurate loudness normalization than if it is used in a single pass.
These instructions use the loudnorm defaults, which align well with PBS recommendations for target loudness. More information can be found at the loudnorm documentation.
Information about PBS loudness standards can be found in the PBS Technical Operating Specifications document. Information about EBU loudness standards can be found in the EBU R 128 recommendation document.
print_format=summary
ffmpeg -i input_file -af aemphasis=type=riaa output_file
This will apply RIAA equalization to an input file allowing correct listening of audio transferred 'flat' (without EQ) from records that used this EQ curve. For more information about RIAA equalization see the Wikipedia page on the subject.
ffmpeg -i input_file -af loudnorm=dual_mono=true -ar 48k output_file
This will normalize the loudness of an input using one pass, which is quicker but less accurate than using two passes. This command uses the loudnorm filter defaults for target loudness. These defaults align well with PBS recommendations, but loudnorm does allow targeting of specific loudness levels. More information can be found at the loudnorm documentation.
ffmpeg -i input_file -af loudnorm=dual_mono=true:measured_I=input_i:measured_TP=input_tp:measured_LRA=input_lra:measured_thresh=input_thresh:offset=target_offset:linear=true -ar 48k output_file
This command allows using the levels calculated using a first pass of the loudnorm filter to more accurately normalize loudness. This command uses the loudnorm filter defaults for target loudness. These defaults align well with PBS recommendations, but loudnorm does allow targeting of specific loudness levels. More information can be found at the loudnorm documentation.
ffmpeg -i input_file -c:v copy -c:a pcm_s16le -af "aresample=async=1000" output_file
ffmpeg -f concat -i mylist.txt -c copy output_file
This command takes two or more files of the same file type and joins them together to make a single file. All that the program needs is a text file with a list specifying the files that should be joined. However, it only works properly if the files to be combined have the exact same codec and technical specifications. Be careful, FFmpeg may appear to have successfully joined two video files with different codecs, but may only bring over the audio from the second file or have other weird behaviors. Don’t use this command for joining files with different codecs and technical specs and always preview your resulting video file!
file './first_file.ext' file './second_file.ext' . . . file './last_file.ext'
-safe 0
ffmpeg -f concat -safe 0 -i mylist.txt -c copy output_file
For more information, see the FFmpeg wiki page on concatenating files.
ffmpeg -i input_1.avi -i input_2.mp4 -filter_complex "[0:v:0][0:a:0][1:v:0][1:a:0]concat=n=2:v=1:a=1[video_out][audio_out]" -map "[video_out]" -map "[audio_out]" output_file
This command takes two or more files of the different file types and joins them together to make a single file.
The input files may differ in many respects - container, codec, chroma subsampling scheme, framerate, etc. However, the above command only works properly if the files to be combined have the same dimensions (e.g., 720x576). Also note that if the input files have different framerates, then the output file will be of variable framerate.
Some aspects of the input files will be normalised: for example, if an input file contains a video track and an audio track that do not have exactly the same duration, the shorter one will be padded. In the case of a shorter video track, the last frame will be repeated in order to cover the missing video; in the case of a shorter audio track, the audio stream will be padded with silence.
0:v:0
0:a:0
0
1
4
If no characteristics of the output files are specified, ffmpeg will use the default encodings associated with the given output file type. To specify the characteristics of the output stream(s), add flags after each -map "[out]" part of the command.
-map "[out]"
For example, to ensure that the video stream of the output file is visually lossless H.264 with a 4:2:0 chroma subsampling scheme, the command above could be amended to include the following: -map "[video_out]" -c:v libx264 -pix_fmt yuv420p -preset veryslow -crf 18
-map "[video_out]" -c:v libx264 -pix_fmt yuv420p -preset veryslow -crf 18
Likewise, to encode the output audio stream as mp3, the command could include the following: -map "[audio_out]" -c:a libmp3lame -dither_method modified_e_weighted -qscale:a 2
-map "[audio_out]" -c:a libmp3lame -dither_method modified_e_weighted -qscale:a 2
To concatenate files of different resolutions, you need to resize the videos to have matching resolutions prior to concatenation. The most basic way to do this is by using a scale filter and giving the dimensions of the file you wish to match:
-vf scale=1920:1080:flags=lanczos
(The Lanczos scaling algorithm is recommended, as it is slower but better than the default bilinear algorithm).
The rescaling should be applied just before the point where the streams to be used in the output file are listed. Select the stream you want to rescale, apply the filter, and assign that to a variable name (rescaled_video in the below example). Then you use this variable name in the list of streams to be concatenated.
rescaled_video
ffmpeg -i input_1.avi -i input_2.mp4 -filter_complex "[0:v:0] scale=1920:1080:flags=lanczos [rescaled_video], [rescaled_video] [0:a:0] [1:v:0] [1:a:0] concat=n=2:v=1:a=1 [video_out] [audio_out]" -map "[video_out]" -map "[audio_out]" output_file
However, this will only have the desired visual output if the inputs have the same aspect ratio. If you wish to concatenate an SD and an HD file, you will also wish to pillarbox the SD file while upscaling. (See the Convert 4:3 to pillarboxed HD command). The full command would look like this:
ffmpeg -i input_1.avi -i input_2.mp4 -filter_complex "[0:v:0] scale=1440:1080:flags=lanczos, pad=1920:1080:(ow-iw)/2:(oh-ih)/2 [to_hd_video], [to_hd_video] [0:a:0] [1:v:0] [1:a:0] concat=n=2:v=1:a=1 [video_out] [audio_out]" -map "[video_out]" -map "[audio_out]" output_file
Here, the first input is an SD file which needs to be upscaled to match the second input, which is 1920x1080. The scale filter enlarges the SD input to the height of the HD frame, keeping the 4:3 aspect ratio; then, the video is pillarboxed within a 1920x1080 frame.
If the input files have different framerates, then the output file may be of variable framerate. To explicitly obtain an output file of constant framerate, you may wish convert an input (or multiple inputs) to a different framerate prior to concatenation.
You can speed up or slow down a file using the fps and atempo filters (see also the Modify speed command).
fps
Here's an example of the full command, in which input_1 is 30fps, input_2 is 25fps, and 25fps is the desired output speed.
ffmpeg -i input_1.avi -i input_2.mp4 -filter_complex "[0:v:0] fps=fps=25 [video_to_25fps]; [0:a:0] atempo=(25/30) [audio_to_25fps]; [video_to_25fps] [audio_to_25fps] [1:v:0] [1:a:0] concat=n=2:v=1:a=1 [video_out] [audio_out]" -map "[video_out]" -map "[audio_out]" output_file
Note that the fps filter will drop or repeat frames as necessary in order to achieve the desired frame rate - see the FFmpeg fps docs for more details.
For more information, see the FFmpeg wiki page on concatenating files of different types.
ffmpeg -i input_file -c copy -map 0 -f segment -segment_time 60 -reset_timestamps 1 output_file-%03d.mkv
Path, name and extension of the output file. In order to have an incrementing number in each segment filename, FFmpeg supports printf-style syntax for a counter.
In this example, '%03d' means: 3-digits, zero-padded Examples:
%03d
%05d
%d
ffmpeg -i input_file -ss 00:02:00 -to 00:55:00 -c copy -map 0 output_file
This command allows you to create an excerpt from a video file without re-encoding the image data.
-ss
-c copy
Variation: trim video by setting duration, by using -t instead of -to
-t
-to
ffmpeg -i input_file -ss 00:05:00 -t 10 -c copy output_file
Note: In order to keep the original timestamps, without trying to sanitise them, you may add the -copyts option.
-copyts
ffmpeg -i input_file -t 5 -c copy -map 0 output_file
This command captures a certain portion of a video file, starting from the beginning and continuing for the amount of time (in seconds) specified in the script. This can be used to create a preview file, or to remove unwanted content from the end of the file. To be more specific, use timecode, such as 00:00:05.
ffmpeg -i input_file -ss 5 -c copy -map 0 output_file
This command copies a video file starting from a specified time, removing the first few seconds from the output. This can be used to create an excerpt, or remove unwanted content from the beginning of a video file.
ffmpeg -sseof -5 -i input_file -c copy -map 0 output_file
This command copies a video file starting from a specified time before the end of the file, removing everything before from the output. This can be used to create an excerpt, or extract content from the end of a video file (e.g. for extracting the closing credits).
ffmpeg -i input_file -c:v libx264 -filter:v "yadif, scale=1440:1080:flags=lanczos, pad=1920:1080:(ow-iw)/2:(oh-ih)/2, format=yuv420p" output_file
yadif=1
Note: the very same scaling filter also downscales a bigger image size into HD.
ffmpeg -i input_file -c:v libx264 -vf "yadif,format=yuv420p" output_file
This command takes an interlaced input file and outputs a deinterlaced H.264 MP4.
"yadif,format=yuv420p" is an FFmpeg filtergraph. Here the filtergraph is made up of one filter chain, which is itself made up of the two filters (separated by the comma). The enclosing quote marks are necessary when you use spaces within the filtergraph, e.g. -vf "yadif, format=yuv420p", and are included above as an example of good practice.
"yadif,format=yuv420p"
-vf "yadif, format=yuv420p"
Note: FFmpeg includes several deinterlacers apart from yadif: bwdif, w3fdif, kerndeint, and nnedi.
For more H.264 encoding options, see the latter section of the encode H.264 command.
Before and after deinterlacing:
The inverse telecine procedure reverses the 3:2 pull down process, restoring 29.97fps interlaced video to the 24fps frame rate of the original film source.
"fieldmatch,yadif,decimate" is an FFmpeg filtergraph. Here the filtergraph is made up of one filter chain, which is itself made up of the three filters (separated by commas). The enclosing quote marks are necessary when you use spaces within the filtergraph, e.g. -vf "fieldmatch, yadif, decimate", and are included above as an example of good practice.
"fieldmatch,yadif,decimate"
Note that if applying an inverse telecine procedure to a 29.97i file, the output framerate will actually be 23.976fps.
This command can also be used to restore other framerates.
Before and after inverse telecine:
ffmpeg -i input_file -c:v video_codec -filter:v setfield=tff output_file
setfield=bff
ffv1
v210
ffmpeg -i input file -filter:v idet -f null -
null
-
E.g For creating access copies with your institutions name
ffmpeg -i input_file -vf drawtext="fontfile=font_path:fontsize=font_size:text=watermark_text:fontcolor=font_colour:alpha=0.4:x=(w-text_w)/2:y=(h-text_h)/2" output_file
fontfile=/Library/Fonts/AppleGothic.ttf
35
text='FFMPROVISR EXAMPLE TEXT'
fontcolor=white
fontcolor=0xFFFFFF
ffmpeg -i input_video file -i input_image_file -filter_complex overlay=main_w-overlay_w-5:5 output_file
main_w-overlay_w-5:5
ffmpeg -i input_file -vf drawtext="fontfile=font_path:fontsize=font_size:timecode=starting_timecode:fontcolor=font_colour:box=1:boxcolor=box_colour:rate=timecode_rate:x=(w-text_w)/2:y=h/1.2" output_file
hh:mm:ss[:;.]ff
timecode='09\\:50\\:01\\:23'
fontcolor=black
fontcolor=0x000000
25/1
Note: -vf is a shortcut for -filter:v.
ffmpeg -i input_file -i subtitles_file -c copy -c:s mov_text output_file
subtitles.srt
mov_text
ASS
SRT
SSA
Note: -c:s is a shortcut for -scodec
-c:s
-scodec
ffmpeg -i input_file -ss 00:00:20 -vframes 1 thumb.png
This command will grab a thumbnail 20 seconds into the video.
ffmpeg -i input_file -vf fps=1/60 out%d.png
This will grab a thumbnail every minute and output sequential png files.
ffmpeg -f image2 -framerate 9 -pattern_type glob -i "input_image_*.jpg" -vf scale=250x250 output_file.gif
This will convert a series of image files into a GIF.
image2
Create high quality GIF
ffmpeg -ss HH:MM:SS -i input_file -filter_complex "fps=10,scale=500:-1:flags=lanczos,palettegen" -t 3 palette.png
ffmpeg -ss HH:MM:SS -i input_file -i palette.png -filter_complex "[0:v]fps=10, scale=500:-1:flags=lanczos[v], [v][1:v]paletteuse" -t 3 -loop 6 output_file
The first command will use the palettegen filter to create a custom palette, then the second command will create the GIF with the paletteuse filter. The result is a high quality GIF.
500:-1
:flags=lanczos
The second command has a slightly different filtergraph, which breaks down as follows:
[0:v]fps=10,scale=500:-1:flags=lanczos[v]
[v][1:v]paletteuse"
paletteuse
Simpler GIF creation
ffmpeg -ss HH:MM:SS -i input_file -vf "fps=10,scale=500:-1" -t 3 -loop 6 output_file
This is a quick and easy method. Dithering is more apparent than the above method using the palette filters, but the file size will be smaller. Perfect for that “legacy” GIF look.
ffmpeg -f image2 -framerate 24 -i input_file_%06d.ext -c:v v210 output_file
-start_number 086400
-i input_file_%06d.ext
ffmpeg -r 1 -loop 1 -i image_file -i audio_file -acodec copy -shortest -vf scale=1280:720 output_file
This command will take an image file (e.g. image.jpg) and an audio file (e.g. audio.mp3) and combine them into a video file that contains the audio track with the image used as the video. It can be useful in a situation where you might want to upload an audio file to a platform like YouTube. You may want to adjust the scaling with -vf to suit your needs.
ffplay -f lavfi "amovie=input_file, asplit=2[out1][a], [a]abitscope=colors=purple|yellow[out0]"
This filter allows visual analysis of the information held in various bit depths of an audio stream. This can aid with identifying when a file that is nominally of a higher bit depth actually has been 'padded' with null information. The provided GIF shows a 16 bit WAV file (left) and then the results of converting that same WAV to 32 bit (right). Note that in the 32 bit version, there is still only information in the first 16 bits.
ffplay -f lavfi "amovie='input.mp3', astats=metadata=1:reset=1, adrawgraph=lavfi.astats.Overall.Peak_level:max=0:min=-30.0:size=700x256:bg=Black[out]"
ffplay -f lavfi "movie='input.mp4', signalstats=out=brng:color=cyan[out]"
ffplay input_file -vf "split=2[m][v], [v]vectorscope=b=0.7:m=color3:g=green[v], [m][v]overlay=x=W-w:y=H-h"
ffmpeg -i input01 -i input02 -filter_complex "[0:v:0]tblend=all_mode=difference128[a];[1:v:0]tblend=all_mode=difference128[b];[a][b]hstack[out]" -map [out] -f nut -c:v rawvideo - | ffplay -
ffprobe -i input_file -show_format -show_streams -show_data -print_format xml
This command extracts technical metadata from a video file and displays it in xml.
See also the FFmpeg documentation on ffprobe for a full list of flags, commands, and options.
ffmpeg -i input_file -map_metadata -1 -c:v copy -c:a copy output_file
Note: -c:v and -c:a are shortcuts for -vcodec and -acodec.
-c:a
-vcodec
-acodec
Bash scripts are plain text files saved with a .sh extension. This entry explains how they work with the example of a bash script named “Rewrap-MXF.sh”, which rewraps .mxf files in a given directory to .mov files.
“Rewrap-MXF.sh” contains the following text:
for file in *.mxf; do ffmpeg -i "$file" -map 0 -c copy "${file%.mxf}.mov"; done
Note: the shell script (.sh file) and all .mxf files to be processed must be contained within the same directory, and the script must be run from that directory. Execute the .sh file with the command sh Rewrap-MXF.sh.
sh Rewrap-MXF.sh
Modify the script as needed to perform different transcodes, or to use with ffprobe. :)
The basic pattern will look similar to this: for item in *.ext; do ffmpeg -i $item (FFmpeg options here) "${item%.ext}_suffix.ext"
for item in *.ext; do ffmpeg -i $item (FFmpeg options here) "${item%.ext}_suffix.ext"
e.g., if an input file is bestmovie002.avi, its output will be bestmovie002_suffix.avi.
Variation: recursively process all MXF files in subdirectories using find instead of for:
find
for
find input_directory -iname "*.mxf" -exec ffmpeg -i {} -map 0 -c copy {}.mov \;
As of Windows 10, it is possible to run Bash via Bash on Ubuntu on Windows, allowing you to use bash scripting. To enable Bash on Windows, see these instructions.
On Windows, the primary native command line programme is PowerShell. PowerShell scripts are plain text files saved with a .ps1 extension. This entry explains how they work with the example of a PowerShell script named “rewrap-mp4.ps1”, which rewraps .mp4 files in a given directory to .mkv files.
“rewrap-mp4.ps1” contains the following text:
$inputfiles = ls *.mp4 foreach ($file in $inputfiles) { $output = [io.path]::ChangeExtension($file, '.mkv') ffmpeg -i $file -map 0 -c copy $output }
$inputfiles
$file
$output
Note: the PowerShell script (.ps1 file) and all .mp4 files to be rewrapped must be contained within the same directory, and the script must be run from that directory.
Execute the .ps1 file by typing .\rewrap-mp4.ps1 in PowerShell.
.\rewrap-mp4.ps1
ffmpeg -i input_file -f null -
This decodes your video and prints any errors or found issues to the screen.
ffmpeg -report -i input_file -f null -
This decodes your video and displays any CRC checksum mismatches. These errors will display in your terminal like this: [ffv1 @ 0x1b04660] CRC mismatch 350FBD8A!at 0.272000 seconds
[ffv1 @ 0x1b04660] CRC mismatch 350FBD8A!at 0.272000 seconds
Frame CRCs are enabled by default in FFV1 Version 3.
-loglevel verbose
ffmpeg -i input_file -f framemd5 -an output_file
This will create an MD5 checksum per video frame.
You may verify an MD5 checksum file created this way by using a Bash script.
ffmpeg -i input_file -af "asetnsamples=n=48000" -f framemd5 -vn output_file
This will create an MD5 checksum for each group of 48000 audio samples. The number of samples per group can be set arbitrarily, but it's good practice to match the samplerate of the media file (so you will get one checksum per second).
Examples for other samplerates:
Note: This filter trandscodes audio to 16 bit PCM by default. The generated framemd5s will represent this value. Validating these framemd5s will require using the same default settings. Alternatively, when your file has another quantisation rates (e.g. 24 bit), then you might add the audio codec -c:a pcm_s24le to the command, for compatibility reasons with other tools, like BWF MetaEdit.
-c:a pcm_s24le
ffmpeg -i input_file -map 0:v:0 -c:v copy -f md5 output_file_1 -map 0:a:0 -c:a copy -f md5 output_file_2
This will create MD5 checksums for the first video and the first audio stream in a file. If only one of these is necessary (for example if used on a WAV file) either part of the command can be excluded to create the desired MD5 only. Use of this kind of checksum enables integrity of the A/V information to be verified independently of any changes to surrounding metadata.
.md5
.txt
Note:The MD5s generated by running this command on WAV files are compatible with those embedded by the BWF MetaEdit tool and can be compared.
ffprobe -f lavfi -i "movie=input_file:s=v+a[in0][in1], [in0]signalstats=stat=tout+vrep+brng, cropdetect=reset=1:round=1, idet=half_life=1, split[a][b];[a]field=top[a1];[b]field=bottom, split[b1][b2];[a1][b1]psnr[c1];[c1][b2]ssim[out0];[in1]ebur128=metadata=1, astats=metadata=1:reset=1:length=0.4[out1]" -show_frames -show_versions -of xml=x=1:q=1 -noprivate | gzip > input_file.qctools.xml.gz
This will create an XML report for use in QCTools for a video file with one video track and one audio track. See also the QCTools documentation.
>
ffprobe -f lavfi -i "movie=input_file,signalstats=stat=tout+vrep+brng, cropdetect=reset=1:round=1, idet=half_life=1, split[a][b];[a]field=top[a1];[b]field=bottom,split[b1][b2];[a1][b1]psnr[c1];[c1][b2]ssim" -show_frames -show_versions -of xml=x=1:q=1 -noprivate | gzip > input_file.qctools.xml.gz
This will create an XML report for use in QCTools for a video file with one video track and NO audio track. See also the QCTools documentation.
ffprobe -f lavfi -i movie=input_file,readeia608 -show_entries frame=pkt_pts_time:frame_tags=lavfi.readeia608.0.line,lavfi.readeia608.0.cc,lavfi.readeia608.1.line,lavfi.readeia608.1.cc -of csv > input_file.csv
This command uses FFmpeg's readeia608 filter to extract the hexadecimal values hidden within EIA-608 (Line 21) Closed Captioning, outputting a csv file. For more information about EIA-608, check out Adobe's Introduction to Closed Captions.
If hex isn't your thing, closed captioning character and code sets can be found in the documentation for SCTools.
Side-by-side video with true EIA-608 captions on the left, zoomed in view of the captions on the right (with hex values represented). To achieve something similar with your own captioned video, try out the EIA608/VITC viewer in QCTools.
ffmpeg -f lavfi -i mandelbrot=size=1280x720:rate=25 -c:v libx264 -t 10 output_file
size
rate
-pix_fmt
yuv420p
ffmpeg -f lavfi -i smptebars=size=720x576:rate=25 -c:v prores -t 10 output_file
ffmpeg -f lavfi -i testsrc=size=720x576:rate=25 -c:v v210 -t 10 output_file
Test an HD video projector by playing the SMPTE colour bars pattern.
ffplay -f lavfi -i smptehdbars=size=1920x1080
Test a VGA (SD) video projector by playing the SMPTE colour bars pattern.
ffplay -f lavfi -i smptebars=size=640x480
Generate a test audio file playing a sine wave.
ffmpeg -f lavfi -i "sine=frequency=1000:sample_rate=48000:duration=5" -c:a pcm_s16le output_file.wav
pcm_s16le
pcm
16
le
Generate a SMPTE bars test video + a 1kHz sine wave as audio testsignal.
ffmpeg -f lavfi -i "smptebars=size=720x576:rate=25" -f lavfi -i "sine=frequency=1000:sample_rate=48000" -c:a pcm_s16le -t 10 -c:v ffv1 output_file
Modifies an existing, functioning file and intentionally breaks it for testing purposes.
ffmpeg -i input_file -bsf noise=1 -c copy output_file
-bsf:v
-bsf:a
Simulates Conway's Game of Life
ffplay -f lavfi life=s=300x200:mold=10:r=60:ratio=0.1:death_color=#C83232:life_color=#00ff00,scale=1200:800
To save a portion of the stream instead of playing it back infinitely, use the following command:
ffmpeg -f lavfi -i life=s=300x200:mold=10:r=60:ratio=0.1:death_color=#C83232:life_color=#00ff00,scale=1200:800 -t 5 output_file
Note: ffmpeg must be compiled with the tesseract library for this script to work (--with-tesseract if using the brew install ffmpeg method).
--with-tesseract
ffplay input_file -vf "ocr,drawtext=fontfile=/Library/Fonts/Andale Mono.ttf:text=%{metadata\\\:lavfi.ocr.text}:fontcolor=white"
Note: FFmpeg must be compiled with the tesseract library for this script to work (--with-tesseract if using the brew install ffmpeg method)
ffprobe -show_entries frame_tags=lavfi.ocr.text -f lavfi -i "movie=input_file,ocr"
ffmpeg -i input_one -i input_two -filter_complex signature=detectmode=full:nb_inputs=2 -f null -
ffmpeg -i input -vf signature=format=xml:filename="output.xml" -an -f null -
Play an image sequence directly as moving images, without having to create a video first.
ffplay -framerate 5 input_file_%06d.ext
Notes:
If -framerate is omitted, the playback speed depends on the images’ file sizes and on the computer’s processing power. It may be rather slow for large image files.
-framerate
You can navigate durationally by clicking within the playback window. Clicking towards the left-hand side of the playback window takes you towards the beginning of the playback sequence; clicking towards the right takes you towards the end of the sequence.
ffmpeg -i input_file -map 0:v:0 video_output_file -map 0:a:0 audio_output_file
This command splits the original input file into a video and audio stream. The -map command identifies which streams are mapped to which file. To ensure that you’re mapping the right streams to the right file, run ffprobe before writing the script to identify which streams are desired.
ffmpeg -i video_file -i audio_file -map 0:v -map 1:a -c copy output_file
This command takes a video file and an audio file as inputs, and creates an output file that combines the video stream in the first file with the audio stream in the second file.
Note: in the example above, the video input file is given prior to the audio input file. However, input files can be added any order, as long as they are indexed correctly when stream mapping with -map. See the entry on stream mapping.
Include the audio tracks from both input files with the following command:
ffmpeg -i video_file -i audio_file -map 0:v -map 0:a -map 1:a -c copy output_file
Create an ISO file that can be used to burn a DVD. Please note, you will have to install dvdauthor. To install dvd author using Homebrew run: brew install dvdauthor
brew install dvdauthor
ffmpeg -i input_file -aspect 4:3 -target ntsc-dvd output_file.mpg
This command will take any file and create an MPEG file that dvdauthor can use to create an ISO.
ffprobe -f lavfi -i movie=input_file,signalstats -show_entries frame=pkt_pts_time:frame_tags=lavfi.signalstats.YDIF -of csv
This ffprobe command prints a CSV correlating timestamps and their YDIF values, useful for determining cuts.
pkt_pts_time
-of
-print_format
ffmpeg -i input_file -filter:v drawbox=w=iw:h=7:y=ih-h:t=max output_file
This command will draw a black box over a small area of the bottom of the frame, which can be used to cover up head switching noise.
ffmpeg -re -i ${INPUTFILE} -map 0 -flags +global_header -vf scale="1280:-1,format=yuv420p" -pix_fmt yuv420p -level 3.1 -vsync passthrough -crf 26 -g 50 -bufsize 3500k -maxrate 1800k -c:v libx264 -c:a aac -b:a 128000 -r:a 44100 -ac 2 -t ${STREAMDURATION} -f tee "[movflags=+faststart]${TARGETFILE}|[f=flv]${STREAMTARGET}"
I use this script to stream to a RTMP target and record the stream locally as .mp4 with only one ffmpeg-instance.
As input, I use bmdcapture which is piped to ffmpeg. But it can also be used with a static videofile as input.
bmdcapture
The input will be scaled to 1280px width, maintaining height. Also the stream will stop after a given time (see -t option.)
${INPUTFILE}
${STREAMDURATION}
${TARGETFILE}
${STREAMTARGET}
ffmpeg -h type=name
encoder=libx264
decoder=mp3
muxer=matroska
demuxer=mov
filter=crop
This section introduces and explains the usage of some additional command line tools similar to FFmpeg for use in digital preservation workflows (and beyond!).
cdda2wav -L0 -t all -cuefile -paranoia paraopts=retries=200,readahead=600,minoverlap=sectors-per-request-1 -verbose-level all output.wav
Cdda2wav is a tool that uses the Paranoia library to facilitate accurate ripping of audio CDs (CDDA). It can be installed via Homebrew with the command brew install cdrtools. This command will accurately rip an audio CD into a single wave file, while querying the CDDB database for track information and creating a cue sheet. This cue sheet can then be used either for playback of the WAV file or to split it into individual access files. Any cdtext information that is discovered will be stored as a sidecar. For more information about cue sheets see this wikipedia article.
brew install cdrtools
-L1
ImageMagick is a free and open-source software suite for displaying, converting, and editing raster image and vector image files.
It's official website can be found here.
Another great resource with lots of supplemental explanations of filters is available at Fred's ImageMagick Scripts.
Unlike many other command line tools, ImageMagick isn't summoned by calling its name. Rather, ImageMagick installs links to several more specific commands: convert, montage, and mogrify, to name a few.
convert
montage
mogrify
compare -metric ae image1.ext image2.ext null:
Compares two images to each other.
Creates thumbnails for all files in a folder and saves them in that folder.
mogrify -resize 80x80 -format jpg -quality 75 -path thumbs *.jpg
montage @list.txt -tile 6x12 -geometry +0+0 output_grid.jpg
convert -verbose input_file.ext | grep -i signature
Gets signature data from an image file, which is a hash that can be used to uniquely identify the image.
mogrify -path ./stripped/ -strip *.jpg
Removes (strips) exif data and moves clean files to a new folder.
convert input_file.ext -resize 750 output_file.ext
This script will also convert the file format, if the output has a different file extension than the input.
The flac tool is the tool created by the FLAC project to transcode to/from FLAC and to manipulate metadata in FLAC files. One advantage it has over other tools used to transcode into FLAC is the capability of embedding foreign metadata (such as BWF metadata). This means that it is possible to compress a BWF file into FLAC and maintain the ability to transcode back into an identical BWF, metadata and all. For a more detailed explanation, see Dave Rice's article on the topic, from which the following commands are adapted.
Use this command to transcode from WAV to FLAC while maintaining BWF metadata
flac --best --keep-foreign-metadata --preserve-modtime --verify input.wav
Use this command to transcode from FLAC to reconstruct original BWF file. Command is the same as the prior command with the exception of substituting --decode for best and changing the input to a .flac file.
--decode
best
.flac
flac --decode --keep-foreign-metadata --preserve-modtime --verify input.flac