How do I concatenate videos, adding a gap between each, using ffmpeg?

[Paddy Landau]

I want something very simple. I want to concatenate a few videos, but leave a brief gap (about half a second) between each.

The videos all have the same format (video dimensions and codecs), and so I don't need any conversion.

To concatenate without a gap is straightforward (video.list is a text file containing a list of the videos in the format file 'video 1.mp4' etc.):

ffmpeg -f concat -i videos.list -codec copy output-video.mp4

Now, how can I do the same but leave a brief gap between each video? The gap doesn't have to be exactly half a second; an approximate value will do.

Thank you

[TheFu]

I'd create a silent/black clip of the same size and length as the others, then I'd use mkvmerge to concatenate the clips in whatever order I needed. If the clips aren't truly using the exact same settings for both audio and video, this won't work. You'll need to transcode.

There's a GUI for mkvmerge in the mkvtools ... I seldom use it. You can also use ffmpeg, but the concatenation with ffmpeg is cumbersome. I have a script that will concatenate sequential files nicely ... er .... it will generate a bash script with mkvmerge commands to be run. Filenames cannot have spaces since inputs are parsed looking for *-01* with either mkv or mp4 extensions. Here's that script:

#!/bin/bash

if [ "X$1" = "X" ] ; then
  echo "Usage: $0 file-01.mkv"
  exit 1;
fi

for filename in "$@"; do

   ROOT=`echo "$filename" | sed -e 's/-01.m[pk][4v]//g'` 
   CNT=1
   if [[ $(echo "$filename" | grep -c 'mkv$') -eq 1 ]] ; then
      EXT="mkv"
   elif  [[ $(echo "$filename" | grep -c 'mp4$') -eq 1 ]] ; then
      EXT="mp4"
   fi
   OUT="ionice mkvmerge -o \"$ROOT.mkv\" \"$ROOT-0$CNT.$EXT\" "
   
   for CNT in {02..99} ; do
     if [ -f "$ROOT-$CNT.$EXT" ] ; then
       OUT="$OUT + \"$ROOT-$CNT.$EXT\""
     fi
   done
   
   echo $OUT 
  #  echo $OUT | sh -
done

If you swap the last 2 echo lines (comment 1, not the other), then you can run the script automatically. I've been doing this for some time. It is safe.

[Paddy Landau]

… mkvmerge

Thank you for this. I installed mkvtoolnix, and then experimented with mkvmerge, using your script as a template.

Unfortunately, it created the same output as ffmpeg did, i.e. without any gaps, but worse, the video was on its side (The videos were all filmed in portrait mode, if that's of any significance.)

I'd create a silent/black clip as the others…

This gives me an idea. I'll create a silent, black clip a half-second long, and insert that between each clip in my ffmpeg command. I'll try that tomorrow, as it's late here and I need to finish up.

First, I just need to figure out how to make that silent, black clip in the right dimensions and with the right encoding!

[TheFu]

First, I just need to figure out how to make that silent, black clip in the right dimensions and with the right encoding!

ffmpeg can take any image and convert it into a clip. So, you just need to take a black image use it 30 or 60 times to get 1 second (30pfs, right?) and have ffmpeg create the detailed resulting clip. To learn the exact encoding parameters - this is very important - I be one of the ffmpeg tools or mplayer -I can dump that information.

Portrait? Eeeew. Nasty. I've never understood why anyone would do that for video. Must be a phone thing. Even in the old days using a Nokia N800, I'd transcode to 240p xvid and watch the videos sideways ... like a 16x9 TV.

[Paddy Landau]

Portrait? Eeeew. Nasty. I've never understood why anyone would do that for video.

I agree! It's a pain, because there is so much extraneous pointless space above and below, while the left and right areas that would have been useful are just black bars. But, alas, this isn't under my control.

My son thinks that it's because people look at their phones in portrait, and instinctively film it that way. I suppose that might be the case.

ffmpeg can take any image and convert it into a clip.

I completely forgot about that!

I did this, which worked, but…

… when playing the concatenated video in VLC, each time the spacer comes around, VLC re-displays the title of the video (as it does right at the start), but with the text distorted through stretching. It's horribly distracting, especially as this is an instruction video with a demonstration that needs to be followed.

The video properties, according to Nautilus, are:

• Dimensions 640×352
• Codec H.264
• Frame rate 30fps
• Bit rate 1838 kbps

You'll notice that the dimensions are given in landscape. I think that the phone saved the videos in landscape, but included metadata to tell the player to rotate the video into portrait.

I created a file, Spacer.png, with the dimensions 352×640. To create the spacer clip of a half of a second, I used this command (I got the command from the internet, because I don't really know what I'm doing):

ffmpeg -loop 1 -i Spacer.png -c:v libx264 -t 0.5 -pix_fmt yuv420p -vf scale=352:640 Spacer.mp4

I created a file, video.list, which contains the list of files as follows:

file 'video 1.mp4'
file 'Spacer.mp4'
file 'video 2.mp4'
file 'Spacer.mp4'
file 'video 3.mp4'
file 'Spacer.mp4'
file 'video 4.mp4'
file 'Spacer.mp4'
file 'video 5.mp4'

This is my command to concatenate the videos:

ffmpeg -f concat -safe 0 -i video.list -codec copy 'Concatenated.mp4'

Do you notice what I did wrong?

[TheFu]

Used ffmpeg, not mkvmerge?
Filenames have spaces?
Not the exact same video encoding parameters as the phone videos in the spacer.

[Paddy Landau]

Used ffmpeg, not mkvmerge?

Because of the problems that I previously mentioned.

Filenames have spaces?

Sure. Why not? (The original names are different, but for privacy reasons, I just used "video 1" and so forth here.)

Not the exact same video encoding parameters as the phone videos in the spacer.

Do you know how I correct that? I am a bit of a dummy when it comes to encoding.

[TheFu]

Because of the problems that I previously mentioned.

Sure. Why not? (The original names are different, but for privacy reasons, I just used "video 1" and so forth here.)

Do you know how I correct that? I am a bit of a dummy when it comes to encoding.

Spaces break everything all the time. Just avoid them or be prepared to run in to dumb issues.
If I didn't encode the files, then I don't have any 100% certain method to know all the settings. It isn't being dumb, it just isn't made easy, since even the version of the encoder might add some unknown thing to the input files that isn't reported in any tool showing the parameters. This is one reason why I think you'll need to re-encode ... so you know the exact parameters for all the videos involved.

[Paddy Landau]

Hmm, alright. Perhaps I'd better just give up trying to add spacers, because it seems like there's a lot of knowledge that I'm lacking.

Thank you for the time that you've spent trying to help!

It's been an interesting experiment, but I'm not much the wiser. I need a "Video encoding for Dummies" book, but I suspect that a dummy wouldn't be able to understand. It's a complex subject, isn't it?

[qyot27]

I'd probably re-encode just to make sure the video is rotated correctly. I'd be willing to bet the issue in VLC is because it may be picking up on some metadata in the segments that changes when it goes from segment into spacer, which throws everything off. When re-encoding, the metadata would [usually] be stripped out, depending on what it is.

And if I did re-encode, it would be by using AviSynth+ to prepare the entire thing before handing it off to FFmpeg.

v1=FFmpegSource2("video1.mp4",atrack=-1).TurnRight()
v2=FFmpegSource2("video2.mp4",atrack=-1).TurnRight()
v3=FFmpegSource2("video3.mp4",atrack=-1).TurnRight()
# [other video files, following the same pattern]
spacer=BlankClip(v1,12)

v1 ++ spacer ++ v2 ++ spacer ++ v3 #[and so on...]

This would be saved as a file, e.g. test_concat.avs. You can then give the script file to FFmpeg for encoding.

The only rub would be the need to build AviSynth+ (easy), either re-build the system FFmpeg with --enable-avisynth (tricky) or build a separate [and preferably static] FFmpeg to handle it (easier, but depends on how many other things you want enabled), and also build the FFMS2 source plugin (easy, if you point it at that other build of FFmpeg). The deb-multimedia (dmo) repository is another option.

One advantage of this approach is that all filtering can be done in the script, and be changed basically on-the-fly so you don't waste time running re-encode after re-encode to get the video processing looking right. If FFmpeg was built with --enable-avisynth, the corresponding FFplay binaries can play the scripts back as if they were normal video files - although the better option would be building mpv and linking against that AviSynth-aware version of FFmpeg, since mpv is actually meant as a normal media player.