Free tool

How an LLM sees your video

ChatGPT, Claude and Gemini can’t watch a video file, but they read images well. Drop in a video and get its contact sheets: a frame every half second, each with its number and time, scene cuts marked, and the sound of that half second (spectrogram, waveform and loudness) drawn under it. Attach them with the index and the model can follow what happens, when, and how it sounds.

Drop a video here, or click to pick oneUp to 5 minutes. It stays on your computer: the sheets are made in your browser.

New MacBook Pro with PROLED XDR Display and Touch by Apple · Watch on YouTube

No video handy?

Try it on New MacBook Pro with PROLED XDR Display and Touch

Apple’s one-minute film for the touchscreen MacBook Pro: fast cuts, bold captions and a driving beat. Watch it, then see its 8 sheets and what Claude makes of them.

What’s on a sheet

16 frames, 8 seconds of video. Each frame is labelled F1, F2… with its time, so you and the model can point at the same moment. A yellow CUT tag marks the first frame of a new shot.

Seeing the sound

Under each frame: a spectrogram (speech is wavy bands, music steady lines, a bleep one flat line), the waveform, and the loudness in dB. Pauses, beats and music beds show up without the model hearing anything.

Your video stays with you

The video is decoded and drawn in your browser and never uploaded, so there’s no wait in a queue. Only if you ask Claude to describe it are the sheets (not the video) sent.