It’s enough of a beast that jellyfin actually created a custom spin of it.

With gstreamer and it's webrtcbin stuff, it's incredibly easy to hack something together that would get an MVP going.

I'd love to see a modular, but integrated approach, and one that lets me treat the media server itself as fungible. I want a sqlite database that is easily syncable/backup/restorable, I want a small Rust component that does media scanning and writes metadata to the database, and another small server component that has "storage" plugins for accessing the actual media and then uses gstreamer to "simul-cast" it to any WebRTC capable client.

Imagine being able to WHIP your media into an OBS session, or have it cast side-by-side with a webcam feed to an RPi hooked up to a TV. Etc.

If I could clone myself, this is the project Clone #1 would work on.

I'd reckon you could get very far relying on GPT-4 to create the appropriate ffmpeg calls based on your precise needs, it has the documentation in its training corpus and is great for this kind of thing.