snapshot: tell ffmpeg pipe input format

[bjia56]

Snapshots of raw images (e.g. via the snapshot url) frequently results in error when rescaling. A sample stacktrace:

[Snapshot Plugin]: -hide_banner -y -i pipe:[REDACTED] -filter_complex scale='min(512,iw)':-2 -frames:v 1 -f image2 pipe:3
[Snapshot Plugin]: pipe:4: Invalid data found when processing input
[Snapshot Plugin]: ffmpeg exited
[Snapshot Plugin]: Snapshot failed Error: ffmpeg input to image conversion failed with exit code: 1, /usr/bin/ffmpeg -hide_banner -y -i pipe:4 -filter_complex scale='min(512,iw)':-2 -frames:v 1 -f image2 pipe:3
[Snapshot Plugin]:     at /src/ffmpeg-image-filter.ts:177:23
[Snapshot Plugin]:     at processTicksAndRejections (node:internal/process/task_queues:95:5)

It looks like this might be a side effect due to switching from static ffmpeg to system ffmpeg (maybe because static ffmpeg is a later version), since static ffmpeg doesn't produce the same error. Explicitly telling system ffmpeg to use jpeg_pipe image2pipe makes it much happier.

[koush]

I’m not familiar with jpeg pipe and whether all ffmpeg has that available. Typically the -f format I use for jpeg is image2. Does that work?

You can actually see the output arguments using that in the logs already.

[bjia56]

image2 doesn't work for the input arguments as it expects a list of files instead of a pipe

[bjia56]

I see there's an image2pipe option, will test that out later today

[koush]

Quick read indicates that the image2* demuxer/muxer handles both jpeg and png, so that's probably preferred.