Analyze a Video with AI

Upload a video and ask what happens in it, and where. Chat Octopus samples frames from the file at known times and reads them one at a time; the spoken audio becomes a transcript with a clock running under every word. Ask "when does the speaker bring up pricing" and the answer comes back with a timestamp on it, not a guess at which fifth of the recording to scrub to.

The shot map

Ask for a rundown of the whole file and what comes back is a shot map: a chronological run of up to about twenty described stills covering the file end to end, whether it runs four minutes or forty. A shot here means the camera holding steady between two cuts, so the map gives one description per cut rather than one per minute. It reads a feature-length upload the same way it reads a short clip, an outline of what happens and in what order, not a frame-by-frame account.

Reading what's on the screen

Those stills get read small, well short of full resolution: enough to know a close-up from a crowd scene, not enough to read a slide's fine print. When the thing you need is on-screen text, name the moment instead of relying on the map. "Read the text on the slide at 14:20" pulls that exact frame at full resolution and reads it directly.

How exact the times are

A time pulled from the picture is accurate to about a second, plenty to find a moment, too loose to place a cut on. Ask for it to snap to something exact instead, the start of a word or the instant a shot changes, both measured rather than guessed at.

The transcript itself comes back as words in order, a start and stop time on each, and nobody's name attached to any of it. Ask a follow-up about what was said and the answer traces back to that line, not a labeled speaker. Before you trust a claim about the picture, ask for the still at the time it names and look at it yourself. The check takes seconds, and it's the difference between a timestamp you use and one you verify.