Natural-Language Search for Video Libraries

Keyword search works when someone has already labeled the thing you're looking for with the exact word you'd search. Video libraries rarely have that — footage gets named by date, camera, or project code, not by what's in the shot. So keyword search on a video library mostly searches metadata nobody wrote carefully.

Instead of matching a keyword to a label, natural-language search matches the meaning of your query to the meaning of what's in the footage. You describe the moment the way you'd describe it to a colleague — "the shot where the founder walks through the office" — and the search works even though nobody ever typed that phrase anywhere.

This only works if the footage has already been indexed for meaning, not just tagged. It's the same underlying capability as semantic video search, applied at the scale of a whole library instead of a single project.

Broll indexes footage for meaning on import, so a plain-language description is enough to find the right moment — no tagging system to maintain, and no need to remember what something was called when it was shot.