How can I remove multiple segments from a video using FFmpeg?

Question

I am trying to delete a few sections of a video using FFmpeg.

For example, imagine if you recorded a show on television and wanted to cut out the commercials. This is simple with a GUI video-editor; you just mark the beginning and ending of each clip to be removed, and select delete. I am trying to do the same thing from the command-line with FFmpeg.

I know how to cut a single segment to a new video like so:

ffmpeg -i input.avi -ss 00:00:20 -t 00:00:05 -map 0 -codec copy output.avi

This cuts a five-second clip and saves it as a new video file, but how can I do the opposite and save the whole video without the specified clip, and how I can I specify multiple clips to be removed?

For example, if my video could be represented by ABCDEFG, I would like to create a new one that would consist of ACDFG.

Answer 1

Well, you still can use the trim filter for that. Here is an example, lets assume that you want to cut out three segments at first and end of the video as well as in the middle:

ffmpeg -i in.ts -filter_complex \
"[0:v]trim=duration=30[a]; \
 [0:v]trim=start=40:end=50,setpts=PTS-STARTPTS[b]; \
 [a][b]concat[c]; \
 [0:v]trim=start=80,setpts=PTS-STARTPTS[d]; \
 [c][d]concat[out1]" -map [out1] out.ts

What I did here? I trimmed first 30 sec, 40-50 sec and 80 sec to end, and then combined them into stream out1 with the concat filter, leaving 30-40 sec (10 sec) and 50-80 sec (30 sec).

About setpts: we need this because trim does not modify picture display time, and when we cut out 10 sec decoder counter does not see any frames for this 10 sec.

If you want to have audio too, You have to do the same for audio streams. So the command should be:

ffmpeg -i utv.ts -filter_complex \
"[0:v]trim=duration=30[av];[0:a]atrim=duration=30[aa];\
 [0:v]trim=start=40:end=50,setpts=PTS-STARTPTS[bv];\
 [0:a]atrim=start=40:end=50,asetpts=PTS-STARTPTS[ba];\
 [av][bv]concat[cv];[aa][ba]concat=v=0:a=1[ca];\
 [0:v]trim=start=80,setpts=PTS-STARTPTS[dv];\
 [0:a]atrim=start=80,asetpts=PTS-STARTPTS[da];\
 [cv][dv]concat[outv];[ca][da]concat=v=0:a=1[outa]" -map [outv] -map [outa] out.ts

Answer 2

I can never get ptQa's solution to work, mostly because I can never figure out what the errors from the filters mean or how to fix them. My solution seems a little clunkier because it can leave behind a mess, but if you're throwing it into a script, the clean up can be automated. I also like this approach because if something goes wrong on step 4, you end up with completed steps 1-3 so recovering from errors is a little more efficient.

The basic strategy is using -t and -ss to get videos of each segment you want, then join together all the parts for your final version.

Say you have 6 segments ABCDEF each 5 seconds long and you want A (0-5 seconds), C (10-15 seconds) and E (20-25 seconds) you'd do this:

ffmpeg -i abcdef.tvshow -t 5 a.tvshow -ss 10 -t 5 c.tvshow -ss 20 -t 5 e.tvshow

or

ffmpeg -i abcdef.tvshow -t 0:00:05 a.tvshow -ss 0:00:10 -t 0:00:05 c.tvshow -ss 0:00:20 -t 0:00:05 e.tvshow

That will make files a.tvshow, c.tvshow and e.tvshow. The -t says how long each clip is, so if c is 30 seconds long you could pass in 30 or 0:00:30. The -ss option says how far to skip into the source video, so it's always relative to the start of the file.

Then once you have a bunch of video files I make a file ace-files.txt like this:

file 'a.tvshow'
file 'c.tvshow'
file 'e.tvshow'

Note the "file" at the beginning and the escaped file name after that.

Then the command:

ffmpeg -f concat -i ace-files.txt -c copy ace.tvshow

That concats all the files in abe-files.txt together, copying their audio and video codecs and makes a file ace.tvshow which should just be sections a, c and e. Then just remember to delete ace-files.txt, a.tvshow, c.tvshow and e.tvshow.

Disclaimer: I have no idea how (in)efficient this is compared to the other approaches in terms of ffmpeg but for my purposes it works better. Hope it helps someone.

Answer 3

For those having trouble following ptQa's approach, there's a slightly more streamlined way to go about it. Rather than concat each step of the way, just do them all at the end.

For each input, define a A/V pair:

//Input1:
[0:v]trim=start=10:end=20,setpts=PTS-STARTPTS,format=yuv420p[0v];
[0:a]atrim=start=10:end=20,asetpts=PTS-STARTPTS[0a];
//Input2:
[0:v]trim=start=30:end=40,setpts=PTS-STARTPTS,format=yuv420p[1v];
[0:a]atrim=start=30:end=40,asetpts=PTS-STARTPTS[1a];
//Input3:
[0:v]trim=start=30:end=40,setpts=PTS-STARTPTS,format=yuv420p[2v];
[0:a]atrim=start=30:end=40,asetpts=PTS-STARTPTS[2a];

Define as many pairs as you need, then concat them all in one pass, where n=total input count.

[0v][0a][1v][1a][2v][2a]concat=n=3:v=1:a=1[outv][outa] -map [outv] -map [outa] out.mp4

This can easily be constructed in a loop.

A complete command that uses 2 inputs might look like this:

ffmpeg -i in.mp4 -filter_complex 
[0:v]trim=start=10.0:end=15.0,setpts=PTS-STARTPTS,format=yuv420p[0v];
[0:a]atrim=start=10.0:end=15.0,asetpts=PTS-STARTPTS[0a];
[0:v]trim=start=65.0:end=70.0,setpts=PTS-STARTPTS,format=yuv420p[1v];
[0:a]atrim=start=65.0:end=70.0,asetpts=PTS-STARTPTS[1a];[0v][0a][1v]
[1a]concat=n=2:v=1:a=1[outv][outa] -map [outv] -map [outa] out.mp4

Answer 4

Edit: This approach will notacibly descync audio and video if the number of segements gets too large (>5-ish). For few segments, it's fine.

This will do it:

 ffmpeg -i input.avi -vf "select='1-between(t,20,25)', setpts=N/FRAME_RATE/TB" -af "aselect='1-between(t,20,25)', asetpts=N/SR/TB" output.avi

This command will discard the segment from 20 seconds to 25 seconds and keep everything else. Instead of an exclusion list, you can also do an inclusion list like this:

ffmpeg -i input.avi -vf "select='lt(t,20)+between(t,790,810)+gt(t,1440)', setpts=N/FRAME_RATE/TB" -af "aselect='lt(t,20)+between(t,790,810)+gt(t,1440)', asetpts=N/SR/TB" output.avi

This will keep 3 segments of the original video (0-20 seconds, 790-810 seconds, and 1440-end seconds), while discarding anything else.

The key parts (of the second example, since it is more complicated):

• -vf is the option for a video filter expression. We use the select filter, which will take an expression, evaluate it for each input frame, and keep the frame if the expression is 1, or discard it if the expression is 0. (There is actually a bit more nuance to this, but it isn't really relevant to your problem. Look here if you want to know the full story.)

• lt(t,20) is the first part of the select filter expression. lt stands for less-than and t is the timestamp of the frame in seconds that is evaluated, so this will evaluate to 1 for any frame with t<20s

• between(t,790,810) will evaluate to 1 for any frame between 790s and 810s

• gt(t, 1440) (greater-than) will become 1 for any frame after 1440s

• We then add all the conditions, to logically or them.

• setpts=N/FRAME_RATE/TB is needed to provide us with the t value that we use in the logical expression. ffmpeg seems to think in samples/frames here, not in seconds, so we use this expression to convert to seconds.

• -af is the option for an audio filter expression. The audio filter works analogous to the video filter, except that to convert to seconds, we use the sample rate SR of the audio, and not the FRAME_RATE of the video.

You can go crazy with the filter expression and add new segments by adding a between(t,...,...) term and combine terms as you need. The exclusion list you wanted simply utilizes the fact that we can invert our filter expression by subtracting it from 1, so everything discarded will now be kept, and vice versa.

Ensure that you have the same select filter for audio and video, or they will get out of sync!

Note that you won't be able to use -codec copy with this solution, since copy tells ffmpeg to not decode the video. Since the video filter is evaluated for decoded frames, we must decode first.

Answer 5

I made a script to speed up editing recorded TV. The script asks you for the beginning and end times of segments you want to keep and splits them out into files. It gives you options, you can:

• Take one or multiple segments.
• You can combine the segments into one resulting file.
• After joining you can keep or delete the part files.
• You can keep the original file or replace it with your new file.

Let me know what you think.

 #!/bin/bash
/bin/date >>segmenter.log

function segment (){
while true; do
    echo "Would you like to cut out a segment ?"
    echo -e "1) Yes\n2) No\n3) Quit"
    read CHOICE
    if [ "$CHOICE" == "3" ]; then
        exit
    elif [ "$CHOICE" == "2" ]; then
        clear
        break
    elif [ "$CHOICE" == "1" ]; then
        clear
        ((segments++))
        echo "What time does segment $segments start ?"
        read SEGSTART
        clear
        echo -e "Segment $segments start set to $SEGSTART\n"  
        echo "What time does segment $segments end ?"
        read SEGEND
        clear
        echo -e "Segment $segments end set to $SEGEND\n"
        break
    else
        clear
        echo -e "Bad option"
        segment "$segments"
    fi
done
if [ "$CHOICE" == "1" ]; then
    echo "Cutting file $file video segment $segments starting at $SEGSTART and ending at $SEGEND"
    ffmpeg -i "$file" -ss $SEGSTART -to  $SEGEND -map 0:0 -map 0:1 -c:a copy -c:v copy  "$filename-part$segments.$extension"  >> segmenter.log 2>&1
    clear
    echo -e "Cut file $filename-part$segments.$extension starting at $SEGSTART and ending at $SEGEND\n"                             
    segment "$segments"
fi
}
           
file="$1"
filename="${file%.*}"
extension="${file##*.}"
clear
segments=0
segment "$segments"
clear
if (("$segments"==1)); then
mv $filename"-part1."$extension "$filename-segmented.$extension"
elif (("$segments">1)); then
echo "Would you like to join the segments into one file ?"      
       OPTIONS="Yes No Quit"
       select opt in $OPTIONS; do
       clear
        if [ "$opt" == "Quit" ]; then
            exit
        elif [ "$opt" == "Yes" ]; then
            clear
            echo "Joining segments"
            ffmpeg -f concat -i <(for f in $filename"-part"*$extension;         do echo "file '$(pwd)/$f'"; done) -c:a copy -c:v copy "$filename-segmented.$extension" >>         segmenter.log 2>&1
            clear
            echo "Would you like to delete the part files ?"
            select opt in $OPTIONS; do
            clear
            if [ "$opt" == "Quit" ]; then
                exit
            elif [ "$opt" == "Yes" ]; then
                for f in $filename"-part"*$extension; do rm $f; done
                break
            elif [ "$opt" == "No" ]; then
                break
            else
                clear
                echo -e "Bad option\n"
            fi
            done
            break
        clear
        elif [ "$opt" == "No" ]; then
            exit
        else
            clear
            echo -e "Bad option\n"
        fi
    done
fi
echo "Would you like to replace the original file with the result of your changes ?"
OPTIONS="Yes No Quit"
select opt in $OPTIONS; do
    clear
    if [ "$opt" == "Quit" ]; then
        exit
    elif [ "$opt" == "Yes" ]; then
        rm $file
        mv "$filename-segmented.$extension" $file
        break
    elif [ "$opt" == "No" ]; then
        break
    else
        clear
        echo -e "Bad option\n"
    fi
done

Answer 6

Converting ABCDEFG to ACDFG is essentially cutting three segments from the video, and then joining those three segments. If all segments will be cut from the same video, and there is no need to change the video's encoding (resolution, bit rate, frame rate, etc.), then re-encoding isn't necessary, and will degrade quality for lossy formats. However, without re-encoding, the cut positions will be not exact but the key frames nearest to the specified times.

The below examples assume that vidfull.mp4 is ABCDEFG, where each letter represents a five minute long segment.

Without re-encoding, the process is almost instantaneous, but cutting needs to be done explicitly, where temporary files are created, and joining can be done either with the concatenation protocol (i.e. file level concatenation), and the .ts container as the intermediate format, which supports concatenation at the file level:

ffmpeg \
  -to 5:0 -i vidfull.mp4 \
  -ss 10:0 -to 20:0 -i vidfull.mp4 \
  -ss 25:0 -i vidfull.mp4 \
  -map 0:v -map 0:a -c copy part1.ts \
  -map 1:v -map 1:a -c copy part2.ts \
  -map 2:v -map 2:a -c copy part3.ts
ffmpeg -i 'concat:part1.ts|part2.ts|part3.ts' -c copy vidcut.ts
ffmpeg -i vidcut.ts -c copy vidcut.mp4
rm -f part{1..3}.ts vidcut.ts

or with the concatenation script demuxer:

ffmpeg \
  -to 5:0 -i vidfull.mp4 \
  -ss 10:0 -to 20:0 -i vidfull.mp4 \
  -ss 25:0 -i vidfull.mp4 \
  -map 0:v -map 0:a -c copy -avoid_negative_ts make_non_negative part1.mp4 \
  -map 1:v -map 1:a -c copy -avoid_negative_ts make_non_negative part2.mp4 \
  -map 2:v -map 2:a -c copy -avoid_negative_ts make_non_negative part3.mp4
cat <<- eof > script.txt
    ffconcat version 1.0
    file '/path to/part1.mp4'
    file '/path to/part2.mp4'
    file '/path to/part3.mp4'
eof
ffmpeg -f concat -safe 0 -i script.txt -c copy vidcut.mp4
rm -f part{1..3}.mp4 script.txt

With re-encoding, the process is slow, but cutting and joining is done with a single command by using the same video as the input to a complex concat filter multiple times after seeking to different positions:

ffmpeg \
  -to 5:0 -i vidfull.mp4 \
  -ss 10:0 -to 20:0 -i vidfull.mp4 \
  -ss 25:0 -i vidfull.mp4 \
  -lavfi concat=n=3:a=1 vidcut.mp4

No need to label input or output pads to the concat filter since there isn't any other filter than concat in the filter chain to use a specific stream from a specific pad, no stream is intended to be excluded from the output, and all streams from unlabelled pads are added to the first output file automatically (the output format needs to support all stream types to avoid a fatal error). And, among the options of the concat filter, n is the number of input segments, and needs to be set to 3 since it is 2 by default, and a is the number of output audio streams, and needs to be set to 1 since it is 0 by default.

Answer 7

Shawnblais’ answer works nicely, but I noticed it was very slow when extracting short fragments from a long video. I suspect ffmpeg decodes the entire video, and applies the trim by discarding frames.

An alternative to using a single input file with multiple trim filters, is to use the same file as input multiple times. Then we can use -ss and -to as input options, which allow for a fast seek. For -ss and -to to be input options, they need to go before the -i that they apply to. The concat filter can then reference them by file number, 0:v, 1:v, etc.

Example concatenating video without audio:

ffmpeg \
  -ss 10.0 -to 15.0 -i in.mkv \
  -ss 50.0 -to 60.0 -i in.mkv \
  -filter_complex '[0:v][1:v]concat=n=2:v=1[outv]' \
  -map '[outv]' \
  out.mkv

Answer 8

Continueing from shawnblais's answer, here's a Python function to help you construct the trim commands given start and end times

def construct_ffmpeg_trim_cmd(timepairs, inpath, outpath):
  cmd = f'ffmpeg -i "{inpath}" -y -filter_complex '
  cmd += '"'
  for i, (start, end) in enumerate(timepairs):
    cmd += (f"[0:v]trim=start={start}:end={end},setpts=PTS-STARTPTS,format=yuv420p[{i}v]; " +
            f"[0:a]atrim=start={start}:end={end},asetpts=PTS-STARTPTS[{i}a]; ")
  for i, (start, end) in enumerate(timepairs):
    cmd += f"[{i}v][{i}a]"
  cmd += f'concat=n={len(timepairs)}:v=1:a=1[outv][outa]'
  cmd += '"'
  cmd += f' -map [outv] -map [outa] "{outpath}"'
  return cmd

GitHub gist: https://gist.github.com/FarisHijazi/eff7a7979440faa84a63657e085ec504

Answer 9

Although the answer provided by ptQa seems to work, I have developed another solution which has proved to work fine.

Essentially, what I do is to cut one video for each part of the original video that I want to include on my result. Later, I concatenate them with the Concat Demuxer explained here.

The result is the same as what I tried first--but which presented sync problems. What I have added is the command -avoid_negative_ts 1 when generating the different videos. With this solution, the sync problems disappear.

Answer 10

Having gone through this using ffmpeg which does a good, if complicated, job I realised ffmpeg seems to be set up well to deal with .mp4 and .ts (transport stream format) files to work losslessly and avoid reencoding.

Looking at dealing with mkv/webm format files which I prefer, I realized there is a great open source package called MKVtoolNix which includes mkvmerge, and as long as you are using a single video file or multiple files with similar formats, it simply defaults to doing the piping and mapping properly, and it accepts very simple syntax to take multiple segments of video and concatenate them perfectly, even supporting chapter information quite easily.

And MKVtoolNix apparently does no transcoding or reencoding so it's inherently very fast and lossless. Output files seem small and optimized. Subtitles and many intricacies of this process are handled by default or with ease. I would keep around ffmpeg for doing a quick cleanup of seek indexing via a "-c copy" command option though.

MKVtoolNix may have some ability to deal with other formats as well, and a GUI version of mkvmerge which looks very impressive.