[FFmpeg-user] hstack with one video offset in time?

[Peter White]

On Fri, Mar 05, 2021 at 11:07:32PM +0100, Paul B Mahol wrote:
> On Fri, Mar 5, 2021 at 10:54 PM Peter White <peter.white at posteo.net> wrote:
> 
> > On Fri, Mar 05, 2021 at 11:33:10AM -0800, Steven Kan wrote:
> > >
> > > >>>
> > > >>> I’d like to assemble these videos, side-by-side, but synced in time,
> > which
> > > >>> means the TrailDown video needs to start 50 seconds after the TrailUp
> > > >>> video. The TrailDown side can be black/blank, or it can be stuck on
> > the
> > > >>> first frame of its video while the right side plays for the first 50
> > > >>> seconds; it doesn’t matter to me.
> > > >>>
> > > >>> I’ve tried all of the following:
> > > >>>
> > > >>> -itsoffset 50 -i TrailDown.mp4 -i TrailUp.mp4
> > > >>> -itsoffset 50 -i TrailDown.mp4 -itsoffset 0 -i TrailUp.mp4
> > > >>> -i TrailDown.mp4 -itsoffset -50 -i TrailUp.mp4
> > > >>>
> > > >>
> > > >> see tpad filter, need recent version.
> > > >
> > > > Thanks! I have figured out the syntax to pad a single video with tpad,
> > e.g.:
> > > >
> > > > ffmpeg -i
> > /Users/steven/Downloads/Record/DownLoad/TrailDown_ch1_20210304010952_20210304011838.mp4
> > -filter_complex "tpad=start_duration=50" tPadOut.mp4
> > > >
> > > > but I’m having trouble with the syntax to delay only one of two videos
> > in an hstack filter:
> > > >
> > > > ffmpeg -i
> > /Users/steven/Downloads/Record/DownLoad/TrailDown_ch1_20210304010952_20210304011838.mp4
> > -i
> > /Users/steven/Downloads/Record/DownLoad/TrailUp_ch1_20210304010838_20210304011506.mp4
> > -filter_complex "tpad=start_duration=50[v0];hstack=inputs=2” Coyote2Up.mp4
> > >
> > >
> > > I brute-forced it by using tpad to pad the first video by itself, and
> > then I used hstack to glue them together. Not the most efficient way of
> > doing things, and I think that induces 2 generations of transcoding, so
> > ideally I’d like to know how to do it properly in the future. But here’s
> > the end result:
> >
> > You need to map your input files to the appropriate filter pads in a
> > complex graph. Since I use this kind of filter graph very rarely, I
> > cannot tell you offhand how to do this properly. The manual section on
> > this should be a good start though.
> >
> > Of course, you can always try the approach I suggested in the other
> > subthread. No transcoding of the original there until you get to the
> > last stage.
> >
> 
> Your approach is very very bad.

Never claimed otherwise, hence the "rough proof of concept". ;)
The question is: Does it give the desired result? Plus, there is no need
for a complex filter graph in that one. As I said, I wouldn't know how
to do that offhand, so a lot of reading might be involved. Obviously the
OP himself is also not familiar with that matter.

I'd take the correct result anytime, even if the way there is not pretty
as opposed to a worse or even no result at all.

Of course, your suggestion is better, no doubt about that.


Peter