Skip to content

Latest commit

 

History

History
526 lines (396 loc) · 36 KB

ffmpeg.md

File metadata and controls

526 lines (396 loc) · 36 KB

Some useful FFmpeg commands

Get FFmpeg from https://ffmpeg.org/download.html.

I currently use the FFmpeg builds from https://www.gyan.dev/ffmpeg/builds/ (for Windows 7+) under the release builds section the file ffmpeg-release-full.7z or directly https://www.gyan.dev/ffmpeg/builds/ffmpeg-release-full.7z

To check your version use

ffmpeg -version

Note: the order of (some) parameters/flags matters - before or after a given input or output (FFmpeg supports multiple input files and creating multiple output streams with one command)

FFplay video viewing

Scroll TOP

Watch a video (looping)

ffplay -v level+error -stats -loop -1 INPUT.mp4

A window will show the video looping infinitly (Q or ESC to exit)

ffplay video controls

Scroll UP | TOP

FFmpeg video editing

Scroll TOP

Only output warnings and stats when running a command

ffmpeg -v level+warning -stats # [...]

Scroll UP | TOP

Convert MKV to MP4

the mkv video file format is suggested when streaming or recording (via OBS) since it can be easily recovert

# Audio codec already is AAC, so it can be copied to save some time
# Also use some compression to shrink the file size a bit
ffmpeg -v level+warning -stats -i INPUT.mkv -c:a copy -c:v libx264 -crf 12 OUTPUT.mp4

Scroll UP | TOP

Convert MP4 to M4A (audio only mp4)

only include audio and subtitles (if present)

ffmpeg -v level+warning -stats -i INPUT.mp4 -c copy -map 0:a -map 0:s? OUTPUT.m4a

or only exclude video

ffmpeg -v level+warning -stats -i INPUT.mp4 -c copy -map 0 -map -0:v OUTPUT.m4a

Scroll UP | TOP

Edit metadata (add chapters)

export all metadata to a file

ffmpeg -v level+warning -stats -i INPUT.mp4 -f ffmetadata FFMETADATAFILE.txt

it looks something like this

;FFMETADATA1

# empty lines or lines starting with ; or # will be ignored
# whitespace will not be ignored so "title = A" would be interpreted as key "title " and value " A"

title=Video Title
artist=Artist Name

# newlines and other special characters like = ; # \ must be escaped with a \
description=Text\
Line two\
\
\
Line five\
Line with Û̕͝͡n̊̑̓̊i͚͚ͬ́c̗͕̈́̀o̵̯ͣ͊ḑ̴̱̐ḛ̯̓̒

# then adding chapters is very simple | order does not matter (no intersection ofc), so easiest is to append them to the end of file

[CHAPTER]
# fractions of a second so 1/1000 says the following START and END are in milliseconds
TIMEBASE=1/1000
# start and end might change a bit when reinserting (snaps to nearest frame when video stream is copied and not encoded)
START=0
END=10000
title=0 to 10sec of the video

[CHAPTER]
TIMEBASE=1/1000
START=10000
END=20000
title=10sec to 20sec of the video

then to reinsert the edited metadata file

ffmpeg -v level+warning -stats -i INPUT.mp4 -i FFMETADATAFILE.txt -map_metadata 1 -c copy OUTPUT.mp4

Scroll UP | TOP

Add thumbnail

ffmpeg -v level+warning -stats -i INPUT.mp4 -i IMAGE.png -map 0 -map 1 -c copy -c:v:1 png -disposition:v:1 attached_pic OUTPUT.mp4

Scroll UP | TOP

Add subtitles

Adding subtitles as an extra stream so they can be turned on and off.

Needs a video player that supports this feature like VLC.

# for mkv output
ffmpeg -v level+warning -stats -i INPUT.mp4 -i SUB.srt -c copy OUTPUT.mkv

# for mp4 output
ffmpeg -v level+warning -stats -i INPUT.mp4 -i SUB.srt -c copy -c:s mov_text OUTPUT.mp4

# ... with multiple subtitle files
ffmpeg -v level+warning -stats -i INPUT.mp4 -i SUB_ENG.srt -i SUB_GER.srt -map 0:0 -map 1:0 -map 2:0 -c copy -c:s mov_text OUTPUT.mp4

# ... with language codes
ffmpeg -v level+warning -stats -i INPUT.mp4 -i SUB_ENG.srt -i SUB_GER.srt -map 0:0 -map 1:0 -map 2:0 -c copy -c:s mov_text -metadata:s:s:0 language=eng -metadata:s:s:1 language=ger OUTPUT.mp4

A subtitle file (.srt) may look like this:

1
00:00:00,000 --> 00:00:03,000
hello there

2
00:00:04,000 --> 00:00:08,000
general kenobi

3
00:00:10,000 --> 00:01:00,000
multi
line
subtitles

displayed like in file → new line in SRT = new line in video

Unicode can be used → tested with z̵̢͎̟͛ͥ̄͑̐͐a̡͈̳̟ͧ̑̓͆̔ͬl̗̠̭͖͓͚ͭ̐͊͊ģ͖͈̍̓ͭͩ̚͝͞ơ̢̞̫̜̞̓͗͊ͪ text and it "pushed" the subtitles of screen (big line height)

Note: not all subtitle files are supported by FFmpeg.

Scroll UP | TOP

Extract frames

# ! subfolder must be created first (recommended)
# dump ALL frames
ffmpeg -v level+warning -stats -i INPUT.mp4 ./_dump/frame%03d.png
# dump frames with custom frame rate (here 1fps)
ffmpeg -v level+warning -stats -i INPUT.mp4 -r 1 ./_dump/frame%03d.png
# dump custom number of frames
ffmpeg -v level+warning -stats -i INPUT.mp4 -frames:v 3 ./_dump/frame%03d.png
# dump all frames in a timeframe (here from 0:00:02 to 0:00:05)
ffmpeg -v level+warning -stats -ss 2 -i INPUT.mp4 -t 3 ./_dump/frame%03d.png
ffmpeg -v level+warning -stats -ss 2 -i INPUT.mp4 -to 5 ./_dump/frame%03d.png
  • png is a good middle ground (lossless compression, but supports less colors)
  • jpeg is slower but has good compression (not lossless compression)
  • bmp is faster but has large file size (uncompressed)

The format frame%03d.png means files will be named: frame000.png, frame001.png, ..., frame050.png, ..., frame1000.png, and so on

-ss, -t, and -to expect a specific time format in short [-][HH:]MM:SS[.m...] or [-]S+[.m...][s|ms|us]

Scroll UP | TOP

Create video from frames

# uses files INPUT000.png, INPUT001.png, etc to create the mp4 video (with 24fps)
ffmpeg -v level+warning -stats -framerate 24 -i INPUT%03d.png OUTPUT.mp4
# uses every png file that starts with INPUT (at 24fps)
ffmpeg -v level+warning -stats -framerate 24 -i INPUT*.png OUTPUT.mp4
# uses every png file (at 24fps)
ffmpeg -v level+warning -stats -framerate 24 -i *.png OUTPUT.mp4

Scroll UP | TOP

crop video

ffmpeg -v level+warning -stats -i INPUT.mp4 -vf crop=WIDTH:HEIGHT:POSX:POSY OUTPUT.mp4

  • WIDTH - the width of the croped window

  • HEIGHT - the height of the croped window

  • POSX - the X position of the croped window (can be omitted = auto center)

  • POSY - the Y position of the croped window (can be omitted = auto center)

  • all values are in pixels, but there is no "px" after it (or an expression that gets calculated each frame)

  • -v documentation

  • -stats documentation

  • full crop filter documentation

Scroll UP | TOP

compress video

lower values are better (higher bitrate), but also lead to larger file size

# for `h.264` values from 18 to 23 are very good
ffmpeg -v level+warning -stats -i INPUT.mp4 -c copy -c:v libx264 -crf 20 OUTPUT.mp4
# for `h.265` values from 24 to 30 are very good
ffmpeg -v level+warning -stats -i INPUT.mp4 -c copy -c:v libx265 -crf 25 OUTPUT.mp4

faster with GPU hardware acceleration / NVIDIA CUDA

# for h.265 → h264_nvenc with NVIDIA CUDA
ffmpeg -v level+warning -stats -hwaccel cuda -hwaccel_output_format cuda -i INPUT.mp4 -c copy -c:v h264_nvenc -fps_mode passthrough -b_ref_mode disabled -preset medium -tune hq -rc vbr -multipass disabled -qp 20 OUTPUT.mp4
# for h.265 → hevc_nvenc with NVIDIA CUDA
ffmpeg -v level+warning -stats -hwaccel cuda -hwaccel_output_format cuda -i INPUT.mp4 -c copy -c:v hevc_nvenc -fps_mode passthrough -b_ref_mode disabled -preset medium -tune hq -rc vbr -multipass disabled -qp 25 OUTPUT.mp4

Scroll UP | TOP

cut video

# start at 0:00:01 and stop at 0:00:10
ffmpeg -v level+warning -stats -ss 1 -i INPUT.mp4 -to 10 -c copy OUTPUT.mp4
# start at 0:00:10 and stop at 0:00:20 (0:00:10 duration)
ffmpeg -v level+warning -stats -ss 10 -i INPUT.mp4 -t 10 -c copy OUTPUT.mp4
# caps output to be 0:00:30 max
ffmpeg -v level+warning -stats -i INPUT.mp4 -t 30 -c copy OUTPUT.mp4

timing from -ss, -to, and -t shift to the nearest frame and not at the exact timestamp when stream is copied (like here)

if exact time is needed the video needs to be re-encoded (-c:v libx264 after or instead of -c copy) which obviously takes longer

when -ss is after -i it will decode and discard the video until the time is reached, when it's before -i like here it will seek into the video without decoding it first (during the seek) so it will be faster.

Scroll UP | TOP

loop video

# loop video infinitely but stop after 0:00:30
ffmpeg -v level+warning -stats -stream_loop -1 -i INPUT.mp4 -t 30 -c copy OUTPUT.mp4
# loop video to length of audio
ffmpeg -v level+warning -stats -stream_loop -1 -i INPUT.mp4 -i INPUT.mp3 -shortest -map 0:v -map 1:a OUTPUT.mp4
# loop audio to length of video
ffmpeg -v level+warning -stats -i INPUT.mp4 -stream_loop -1 -i INPUT.mp3 -shortest -map 0:v -map 1:a OUTPUT.mp4

if exact timing is needed, it is better to re-encode the video (-c:v libx264 after or instead of -c copy)

Scroll UP | TOP

Reverse video and/or Audio

Warning: these filters require a lot of memory (buffer of the entire clip) so it's suggested to also use the trim filter as shown

# reverse video only (first 5sec)
ffmpeg -v level+warning -stats -i INPUT.mp4 -vf trim=end=5,reverse OUTPUT.mp4
# reverse audio only (first 5sec)
ffmpeg -v level+warning -stats -i INPUT.mp4 -af atrim=end=5,areverse OUTPUT.mp4

Scroll UP | TOP

Concatenate multiple videos into one

# using filter complex and the concat filter (if video formats are not the same add `:unsafe` to the `concat` filter)
ffmpeg -v level+warning -stats -i INPUT_0.mp4 -i INPUT_1.mp4 -filter_complex "[0:v] [0:a] [1:v] [1:a] concat=n=2:v=1:a=1 [v1] [a1]" -map "[v1]" -map "[a1]" OUTPUT.mp4
# using a list file and demuxer
ffmpeg -v level+warning -stats -safe 0 -f concat -i VIDEO_LIST.txt -c copy OUTPUT.mp4

content of VIDEO_LIST.txt as follows

file 'INPUT_0.mp4'
file 'INPUT_1.mp4'

Scroll UP | TOP

Cut and combine multiple sections of multiple files

Cut clips and concat them (with re-encoding) as follows (video and audio are cut and combined separately).

# 00:00 to 00:02 video and audio of INPUT_0.mp4
# 00:04 to 00:08 video and audio of INPUT_0.mp4
# 00:01 to 00:05 video and audio of INPUT_1.mp4
# 00:06 to 00:08 video and audio of INPUT_1.mp4
ffmpeg -v level+warning -stats -i INPUT_0.mp4 -i INPUT_1.mp4 -filter_complex "[0:v]trim=0:2,setpts=PTS-STARTPTS[i0v0];[0:v]atrim=0:2,asetpts=PTS-STARTPTS[i0a0];[0:v]trim=4:8,setpts=PTS-STARTPTS[i0v1];[0:v]atrim=4:8,asetpts=PTS-STARTPTS[i0a1];[1:v]trim=1:5,setpts=PTS-STARTPTS[i1v0];[1:v]atrim=1:5,asetpts=PTS-STARTPTS[i1a0];[1:v]trim=6:8,setpts=PTS-STARTPTS[i1v1];[1:v]atrim=6:8,asetpts=PTS-STARTPTS[i1a1];[i0v0][i0a0][i0v1][i0a1][i1v0][i1a0][i1v1][i1a1]concat=n=4:v=1:a=1[cv][ca]" -map "[cv]" -map "[ca]" OUTPUT.mp4
# with (h.264) NVIDIA:CUDA and slow/low compression (4 to 8 MB variable bitrate and 4 QP) for the first video stream of the combined video clips
ffmpeg -v level+warning -stats -hwaccel cuda -hwaccel_output_format cuda -i INPUT_0.mp4 -hwaccel cuda -hwaccel_output_format cuda -i INPUT_1.mp4 -filter_complex "[0:v]trim=0:2,setpts=PTS-STARTPTS[i0v0];[0:v]atrim=0:2,asetpts=PTS-STARTPTS[i0a0];[0:v]trim=4:8,setpts=PTS-STARTPTS[i0v1];[0:v]atrim=4:8,asetpts=PTS-STARTPTS[i0a1];[1:v]trim=1:5,setpts=PTS-STARTPTS[i1v0];[1:v]atrim=1:5,asetpts=PTS-STARTPTS[i1a0];[1:v]trim=6:8,setpts=PTS-STARTPTS[i1v1];[1:v]atrim=6:8,asetpts=PTS-STARTPTS[i1a1];[i0v0][i0a0][i0v1][i0a1][i1v0][i1a0][i1v1][i1a1]concat=n=4:v=1:a=1[cv][ca]" -map "[cv]" -c:v:0 h264_nvenc -preset p7 -tune hq -profile:v:0 high -level:v:0 auto -rc vbr -b:v:0 4M -minrate:v:0 500k -maxrate:v:0 8M -bufsize:v:0 8M -multipass disabled -fps_mode passthrough -b_ref_mode:v:0 disabled -rc-lookahead:v:0 32 -qp 4 -map "[ca]" OUTPUT.mp4
# the `-hwaccel cuda -hwaccel_output_format cuda` must be in front of every input video (that is in the filter and gets encoded as video stream)

Scroll UP | TOP

Create/download video with m3u8 playlist

# this will whitelist urls (`-i`) for files available via file, http/s, tcp, tls, or crypto protocol (for this command, not permanent)
ffmpeg -v level+warning -stats -protocol_whitelist file,http,https,tcp,tls,crypto -i INPUT.m3u8 -c copy OUTPUT.mp4
ffmpeg -v level+warning -stats -protocol_whitelist file,http,https,tcp,tls,crypto -i https://example.com/INPUT.m3u8 -c copy OUTPUT.mp4

Scroll UP | TOP

find silence parts in video

# finds sections min 240sec long and max -70db loud and writes them to LOG.txt
ffmpeg -v level+warning -stats -i INPUT.mp4 -af silencedetect=noise=-70dB:d=240 -f null - 2> LOG.txt

look for [silencedetect @ * lines in log file

[silencedetect @ 0000000000******] silence_start: 01:00:02.500
[silencedetect @ 0000000000******] silence_end: 01:10:02.500 | silence_duration: 00:09:59.989
[silencedetect @ 000000000*******] silence_start: 02:00:02.500
[silencedetect @ 000000000*******] silence_end: 02:10:02.500 | silence_duration: 00:09:59.989
[...]

Scroll UP | TOP