Canvas Film

You cannot hear the film, and a still is not the film. Every step here exists to replace a sense you lack: speech-to-text is your ears, stills and contact sheets are your eyes, and word timestamps are your sense of timing. Ask of every beat: which word does this land on, and have I checked it landed?

Two rules carry the rest:

1. Audio first, then pictures. Lock the voice timeline before animating. Every visual beat is anchored to a word (word("c02", "bail")), never to a typed number, so re-voicing a line moves the gag with it.
2. One pure renderAt(t). The page draws any moment from t alone. Live playback passes audio.currentTime; export steps t = frame / fps. Nothing reads the wall clock, so what you review is what renders.

Pipeline