content understanding and B-roll search for agent-built films

[bxaiqiyi]

Hi —

I build films with OpenFilm through an agent, and the one step that still needs a human is deciding what’s on screen and finding extra visuals to cover it.

Could OpenFilm expose an API-level way to (1) understand a recording — its shots, what each shot shows, where the speaker is talking, and word timings — and (2) search for B-roll and place candidates onto the timeline? Ideally an agent could query both and get structured results, the same way we already use get for music and transitions.

That would let an agent cut a long talking-head recording and cover it with B-roll without the person picking clips by hand. Even the understanding part alone would be a big unlock.

Thanks!

[bxaiqiyi]

Of course, I’d rather the system provide this capability out of the box — but I’m not able to build these features myself.

[tobicon]

Mate, the product is becoming a monster, you don't need all features, u need tools that decrease the time to a 90% polished post production footage

[Sadrak2017]

Would a read-only shot index be a useful first slice, returning source clip IDs, time ranges and descriptions, with candidate B-roll separate from timeline edits? An agent could compare candidates before proposing a change the person can review. Our work on Sinaps Studio at SinapsData combines model comparison, data orchestration and agent monitoring; keeping analysis evidence separate from committed edits seems useful here too.