Option -ss on output with -c copy uses dts instead of pts

[Misaki]

Editing scenario: have a finished video, want to copy audio from another video starting at a certain point.

If you use the command

ffmpeg -ss 11 -i video1.webm -i video2.mp4 -c copy output.mkv

then ffmpeg will seek to the video keyframe before 11 in video1.webm, and copy audio from there, instead of from 11.

If you add -ss 0 to the output, it trims audio before 11. But due to this bug, if the first dts of video2.mp4 is negative, it also trims the first group of pictures.

How to replicate:

ffmpeg -filter_complex color=black -t 2 -g 25 black.mp4; ffmpeg -i black.mp4 -c copy -ss 0 black2.mp4; ffmpeg -i black.mp4 -ss 0 black3.mp4

The first file is 50 frames. The second, using -c copy, is only 25 frames. The third, not copying, is 50 frames. Verbose output for these commands attached to report.

The desired behaviour is that packets are trimmed based on their presentation timestamp (pts), not their decode timestamp (dts).

[Misaki]

First command's verbose output

[Misaki]

Second command's verbose output

[Misaki]

Third command's verbose output

[Gyan]

For your root use case, you can run

ffmpeg -seek2any 1 -ss 11 -i video1.webm -i video2.mp4 -c copy -map 1:v -map 0:a output.mkv

For supporting demuxers (like Matroska), this won't seek to an earlier KF.

[Misaki]

Thanks