A search engine for what's on screen in a video — action, framing, who's in shot — rather than just the transcript.
If you've ever edited video you know how this goes. You're hunting for one specific shot you remember seeing, and you end up scrubbing the timeline back and forth to find it.
You upload the video, type what you're looking for in plain English, and get the matching clip back. An example query, run against the Django Unchained demo: "man in blue dress riding a horse with him in the center of the frame" — and it returns the matching scene.
Most video-search tools work off the transcript, which is only what someone said out loud. A lot of what's useful to find in a video isn't spoken though — it's visual.
The upload button is off for now — processing a video takes storage and compute, and I'm not paying for that yet. The films already loaded in are fully searchable.
Two things I want to figure out:
→ Which use cases are actually worth building on — both as a standalone app and as an API other tools could use.
→ Where the current system breaks down and what its real limitations are.
If you try it and something doesn't work, or something does and surprises you, let me know. krithikssg@gmail.com.