Frame Bench — video to text

01 Endpoint

API key Base URL Model Request shape

02 Source

File — mp4, mov, webm, jpg, png Sampling

Interval (seconds)

Max frames Long edge px Detail

03 Prompt

$ / 1M in $ / 1M out

Your key goes straight from this browser tab to the base URL above and nowhere else. Serve this page over http://localhost — opening it as a file:// URL will fail the CORS preflight.

Ready.

Frames held

Est. image tok

—

Input tok

—

Output tok

—

Cost

—

Latency

—

Model output

Nothing sent yet.

How the sampling works

Frames come out of a hidden <video> element: the page seeks to a timestamp, paints that instant onto a canvas, and encodes it as JPEG. The browser exposes time, not frame numbers, so every Nth frame is converted to an interval of N ÷ fps seconds — if you enter the wrong source fps, the spacing shifts but nothing breaks.

These are evenly spaced samples, not true I-frames. Reading real keyframes needs WebCodecs plus an MP4 demuxer. For describing a clip, even spacing is usually the better sample anyway.

The image token estimate uses the classic 85 + 170-per-512px-tile tiling rule and ignores your text prompt, so treat it as a floor for budgeting. The counts reported after a call are the real ones from the API.

Decoding is done by the browser, so codec support is whatever your browser has. H.264 MP4 is safe everywhere; HEVC and some AV1 files will refuse to load.