176

There is often a need in the open source or active developer community to publish large video segments online. (Meet-up videos, campouts, tech talks...) Being that I am a developer and not a videographer I have no desire to fork out the extra scratch on a premium Vimeo account. How then do I take a 12.5 GB (1:20:00) MPEG tech talk video and slice it into 00:10:00 segments for easy uploading to video sharing sites?

Braiam
  • 35,991
Gabriel
  • 1,863

11 Answers11

197

Here is the one line solution:

ffmpeg -i input.mp4 -c copy -map 0 -segment_time 00:20:00 -f segment output%03d.mp4

Please note that this does not give you accurate splits, but should fit your needs. It will instead cut at the first frame after the time specified after segment_time, in the code above it would be after the 20 minute mark.

If you find that only the first chunk is playable, try adding -reset_timestamps 1 as mentioned in the comments.

ffmpeg -i input.mp4 -c copy -map 0 -segment_time 00:20:00 -f segment -reset_timestamps 1 output%03d.mp4
Jon
  • 2,207
  • This is the best answer, very elegant way. Does the job, no additional scripts – Abhijeet Apsunde Jun 08 '16 at 07:06
  • 14
    It actually gives you very accurate splits, if you value video quality. Rather than splitting based on a particular time, it splits on the nearest keyframe following the requested time, so each new segment always starts with a keyframe. – Malvineous Feb 25 '17 at 06:51
  • 4
    what are the units? 8s? 8min? 8h? – user1133275 Mar 20 '17 at 20:58
  • 1
    @user1133275 its second – Jon Mar 20 '17 at 21:11
  • actually this should be marked as the answer – Illarion Kovalchuk Apr 19 '17 at 13:25
  • 1
    @user1133275 you can use -segment_time 00:20:00 notation instead – Mahmoud Mostafa Apr 25 '17 at 16:11
  • 6
    On Mac, I found that this resulted in N output video chunks but only the 1st of them was a valid, viewable MP4. The other N-1 chunks were blank video (all black) with no audio. To make it work, I needed to add the reset_timestamps flag like so: ffmpeg -i input.mp4 -c copy -map 0 -segment_time 8 -f segment -reset_timestamps 1 output%03d.mp4. – jarmod Jun 05 '17 at 15:32
  • Since the splits are not accurate (from the perspective of time), is it possible for ffmpeg to also output the frame number of each split? – hoj201 Sep 25 '17 at 18:21
  • When I run this, I get the following error after the first segment is processed: [avi @ 05164500] Too large number of skipped frames 216000 > 60000A speed= 782x x av_interleaved_write_frame(): Invalid argument – jlarsch Oct 30 '17 at 07:49
  • 27
    found that adding -reset_timestamps 1 fixes the issue for me – jlarsch Oct 30 '17 at 08:27
  • I also found that -reset_timestamps 1 was needed for me on Ubuntu, so I'm guessing it has to do with the specific format or codec? – Jonathan Nov 19 '17 at 07:39
  • Sir, this works well for me. splits .mp3 also into chunks of required lengths say 1 second each. what should I do , if I wanted to split chunks in successive series. i.e first iteration will be 0 -> 1, 1->2 , 2->3 etc. 2nd iteration will be 0.5->1.5, 1.5 to 2.5 and so on. – kRazzy R Dec 06 '17 at 16:39
  • This is the first I've heard of a way to specifically get ffmpeg to split at keyframes! Usually it tries split cut the video at the exact time I specify, which if I'm using -codec copy results in a second-or-so of broken video. – Wowfunhappy Sep 30 '19 at 13:46
  • This is the correct answer. – Jake Ireland Sep 20 '20 at 09:52
  • Using this answer compared to the answer by @warren-young I ended up with a faster run, and for some reason smaller segment files, even though I used the -c copy on both versions. – aVeRTRAC Oct 18 '20 at 18:03
151
$ ffmpeg -i source-file.foo -ss    0 -t 600  first-10-min.m4v
$ ffmpeg -i source-file.foo -ss  600 -t 600 second-10-min.m4v
$ ffmpeg -i source-file.foo -ss 1200 -t 600  third-10-min.m4v
...

Wrapping this up into a script to do it in a loop wouldn't be hard.

Beware that if you try to calculate the number of iterations based on the duration output from an ffprobe call that this is estimated from the average bit rate at the start of the clip and the clip's file size unless you give the -count_frames argument, which slows its operation considerably.

Another thing to be aware of is that the position of the -ss option on the command line matters. Where I have it now is slow but accurate. The linked article describes fast-but-inaccurate and slower-but-still-accurate alternative formulations. You pay for the latter with a certain complexity.

All that aside, I don't think you really want to be cutting at exactly 10 minutes for each clip. That will put cuts right in the middle of sentences, even words. I think you should be using a video editor or player to find natural cut points just shy of 10 minutes apart.

Assuming your file is in a format that YouTube can accept directly, you don't have to reencode to get segments. Just pass the natural cut point offsets to ffmpeg, telling it to pass the encoded A/V through untouched by using the "copy" codec:

$ ffmpeg -i source.m4v -ss       0 -t  593.3 -c copy part1.m4v
$ ffmpeg -i source.m4v -ss   593.3 -t 551.64 -c copy part2.m4v
$ ffmpeg -i source.m4v -ss 1144.94 -t 581.25 -c copy part3.m4v
...

The -c copy argument tells it to copy all input streams (audio, video, and potentially others, such as subtitles) into the output as-is. For simple A/V programs, it is equivalent to the more verbose flags -c:v copy -c:a copy or the old-style flags -vcodec copy -acodec copy. You would use the more verbose style when you want to copy only one of the streams, but re-encode the other. For example, many years ago there was a common practice with QuickTime files to compress the video with H.264 video but leave the audio as uncompressed PCM; if you ran across such a file today, you could modernize it with -c:v copy -c:a aac to reprocess just the audio stream, leaving the video untouched.

The start point for every command above after the first is the previous command's start point plus the previous command's duration.

Warren Young
  • 72,032
21

Just use what is built into ffmpeg to do exactly this.

ffmpeg -i invid.mp4 -threads 3 \
       -vcodec copy -f segment -segment_time 10:00 \
       -reset_timestamps 1 \
       cam_out_h264_%02d.mp4

This will split it into roughly 10-minute chunks, split at the relevant keyframes, and will output to the files cam_out_h264_01.mp4, cam_out_h264_02.mp4, etc.

John Allard
  • 1,368
20

An Alternate more readable way would be

ffmpeg -i input.mp4 -ss 00:00:00 -to 00:10:00 -c copy output1.mp4
ffmpeg -i input.mp4 -ss 00:10:00 -to 00:20:00 -c copy output2.mp4

/**
* -i  input file
* -ss start time in seconds or in hh:mm:ss
* -to end time in seconds or in hh:mm:ss
* -c codec to use
*/

Here's the source and list of Commonly used FFmpeg commands.

13

If you want to create really same Chunks must force ffmpeg to create i-frame on the every chunks' first frame so you can use this command for create 0.5 second chunk.

ffmpeg -hide_banner  -err_detect ignore_err -i input.mp4 -r 24 -codec:v libx264  -vsync 1  -codec:a aac  -ac 2  -ar 48k  -f segment   -preset fast  -segment_format mpegts  -segment_time 0.5 -force_key_frames  "expr: gte(t, n_forced * 0.5)" out%d.mkv
  • This should be the accepted answer. Thanks for sharing mate! – Stephan Ahlf Jul 21 '19 at 07:02
  • 1
    @StephanAhlf: Usually you want to avoid re-encoding, both for quality and CPU usage. It's only worth doing this if you really need exact split points and there aren't already I-frames there in the original. This also has a bunch of arbitrary choices for options, like forcing a frame-rate of exactly 24 FPS, not 24000/1001, and libx264's defaults of CRF23, plus a lower-quality preset fast, and FFmpeg's low-quality aac built-in encoder with 48k bitrate. – Peter Cordes Jan 08 '22 at 15:12
  • there's to much in this command without explanation for being a good answer – Daniel Alder Jun 11 '22 at 23:20
10

Faced the same problem earlier and put together a simple Python script to do just that (using FFMpeg). Available here: https://github.com/c0decracker/video-splitter, and pasted below:

#!/usr/bin/env python
import subprocess
import re
import math
from optparse import OptionParser
length_regexp = 'Duration: (\d{2}):(\d{2}):(\d{2})\.\d+,'
re_length = re.compile(length_regexp)
def main():
    (filename, split_length) = parse_options()
    if split_length <= 0:
        print "Split length can't be 0"
        raise SystemExit
    output = subprocess.Popen("ffmpeg -i '"+filename+"' 2>&1 | grep 'Duration'",
                              shell = True,
                              stdout = subprocess.PIPE
    ).stdout.read()
    print output
    matches = re_length.search(output)
    if matches:
        video_length = int(matches.group(1)) * 3600 + \
                       int(matches.group(2)) * 60 + \
                       int(matches.group(3))
        print "Video length in seconds: "+str(video_length)
    else:
        print "Can't determine video length."
        raise SystemExit
    split_count = int(math.ceil(video_length/float(split_length)))
    if(split_count == 1):
        print "Video length is less then the target split length."
        raise SystemExit
    split_cmd = "ffmpeg -i '"+filename+"' -vcodec copy "
    for n in range(0, split_count):
        split_str = ""
        if n == 0:
            split_start = 0
        else:
            split_start = split_length * n
            split_str += " -ss "+str(split_start)+" -t "+str(split_length) + \
                         " '"+filename[:-4] + "-" + str(n) + "." + filename[-3:] + \
                         "'"
    print "About to run: "+split_cmd+split_str
    output = subprocess.Popen(split_cmd+split_str, shell = True, stdout =
                              subprocess.PIPE).stdout.read()
def parse_options():
    parser = OptionParser()
    parser.add_option("-f", "--file",
                      dest = "filename",
                      help = "file to split, for example sample.avi",
                      type = "string",
                      action = "store"
    )
    parser.add_option("-s", "--split-size",
                      dest = "split_size",
                      help = "split or chunk size in seconds, for example 10",
                      type = "int",
                      action = "store"
    )
    (options, args) = parser.parse_args()
    if options.filename and options.split_size:
        return (options.filename, options.split_size)
    else:
        parser.print_help()
        raise SystemExit
if __name__ == '__main__':
    try:
        main()
    except Exception, e:
        print "Exception occured running main():"
        print str(e)
  • 1
    Next time do this please in a comment. Link-only answers aren't really liked here, and the same if you advert your site. If it is an opensource project with source code, maybe it is an exception, but I now risked my reviewing privileges by not voting for the removal of your answer. And yes, you can't post comments, but after you collected 5 upvotes (which seems very fast in your case) you will. – peterh Nov 21 '14 at 00:40
  • 3
    Hi and welcome to the site. Please don't post link only answers. While your script itself would probably make a great answer, a link to it is not an answer. It is a signpost pointing to an answer. More on that here. Since you kindly gave the link, I went ahead and included the script in the body of your answer. If you object to that, please delete the answer altogether. – terdon Nov 21 '14 at 01:54
  • @terdon, you probably shouldn't have done that without specific permission to redistribute as Creative Commons. The source has the Apache licence, which doesn't automatically confer that right. – Toby Speight Jul 12 '21 at 06:46
  • @TobySpeight the OP is the author of the tool. If they objected, they would have removed it and without the code, the answer would have been deleted. – terdon Jul 12 '21 at 08:29
  • Yes, but you'd still have infringed the licence. In this case, it appears you got away with that. – Toby Speight Jul 12 '21 at 10:27
  • Thank you so much! This was incredibly helpful. – Jeel Shah Jan 16 '22 at 16:54
4

Note the exact punctuation of the alternative format is -ss mm:ss.xxx. I struggled for hours trying to use the intuitive-but-wrong mm:ss:xx to no avail.

$ man ffmpeg | grep -C1 position

-ss position
Seek to given time position in seconds. "hh:mm:ss[.xxx]" syntax is also supported.

References here and here.

Caleb
  • 70,105
1
ffmpeg -i input.mp4 -force_key_frames expr:gte(t,n_forced*600) -f segment -segment_time 600 -reset_timestamps 1 -map 0 -segment_format_options movflags=+faststart output_%03d.mp4

To split the video more accurately, you can use the -force_key_frames option to specify keyframes at the desired split points.

  • -f segment -segment_time 600 sets the segment length
  • gte(t, n_forced*600): This is the expression itself, where:
    • t represents the current time of the video frame being processed.
    • n_forced represents the number of keyframes that have been forced so far.
    • *600 multiplies n_forced by 600, which is the time interval in seconds.
Yashik
  • 111
0
#!/bin/bash

if [ "X$1" == "X" ]; then
    echo "No file name for split, exiting ..."
    exit 1
fi

if [ ! -f "$1" ]; then
    echo "The file '$1' doesn't exist. exiting ..."
    exit 1
fi

duration=$(ffmpeg -i "$1" 2>&1 | grep Duration | sed 's/^.*Duration: \(.*\)\..., start.*$/\1/' | awk -F: '{ print ($1 * 3600) + ($2 * 60) + $3 }')      #'
split_time=${split_time:-55}
time=0
part=1

filename=${file%%.*}
postfix=${file##*.}

while [ ${time} -le ${duration} ]; do

echo    ffmpeg -i "$1" -vcodec copy -ss ${time} -t ${split_time} "${filename}-${part}.${postfix}"
    (( part++ ))
    (( time = time + split_time ))

done
0

One line solution


for i in {3..100}; do ffmpeg -i input.mp4 -ss $(($((i))*600)) -t $(($((i+1))*600)) $i.mp4 ; done;

Trect
  • 109
  • 2
    Welcome to the site, and thank your for your contribution. Please note that it is not necessary to have nested $((...)) constructs, $(( i*600 )) or $(( (i+1)*600 )) are totally fine. Also, it is good practice to double-quote parameter expansions, as in "$i.mp4" (although in your example there is little danger of unwanted word-splitting). – AdminBee Sep 24 '20 at 07:13
0

Please have a look into below command

fun trim(): Array<String?>{
        val cmdList = ArrayList<String>()
        var cmds: Array<String?>? = null
        try{
            cmdList.add("-y")
        cmdList.add(&quot;-ss&quot;)
        cmdList.add(startduration.toString())
        cmdList.add(&quot;-t&quot;)
        cmdList.add(endduration.toString())
        cmdList.add(&quot;-i&quot;)
        cmdList.add(sourcevideopath)
        cmdList.add(&quot;-vcodec&quot;)
        cmdList.add(&quot;copy&quot;)

        cmdList.add(&quot;-movflags&quot;)
        cmdList.add(&quot;faststart&quot;)
        cmdList.add(&quot;-preset&quot;)
        cmdList.add(&quot;ultrafast&quot;)

        cmdList.add(outputvideopath)

        cmds = arrayOfNulls&lt;String&gt;(cmdList.size)
        cmdList.toArray(cmds)

    }catch (e:Exception){
        e.printStackTrace()
    }
    return cmds!!
}

  • 2
    Welcome to the site, and thank you for your contribution. You may want to add some explanation on what this command does and how to use it to solve the OPs problem. – AdminBee Jun 23 '21 at 11:25
  • 2
    What language is that, and why did you choose it instead of an ordinary shell for running commands? – Toby Speight Jun 24 '21 at 10:29
  • @TobySpeight looks like Kotlin, and it's unusual because all it does is create an argv for launching ffmpeg and doesn't launch it itself in a subprocess or other such mechanism – moshbear Jul 12 '21 at 06:28