We made a film with code, for free, on one laptop
A 60-second film about one failed experiment, drawn frame by frame by code and narrated by a free open voice. Press play, then see how we planned it, checked it and what we still don't know.
Sandbox
Under the hood
What you are watching. A 60-second film, drawn frame by frame by code, with a voice that is a free open model. It tells one failed experiment: we gave a small AI that holds about twenty minutes of talking a notebook, and it did not help (7 of 24 right, 8 without), because what it wrote down was the weak step. Handed the exact paragraph, it got 15 of 15 simple questions. Finding that paragraph is next. One run, scored by hand.
How it was made. The script comes first and goes on a one-page review board (open the example board) that shows each line, a rough sketch of what you will see, and checks (length, pace, numbers traced to their source, jargon) in about a second, before any picture is drawn. Then one command builds the voice, the pictures, the sound and the film from that script. A director skill holds the method and a checker.
Honest results. The checker passed 14 of 16: the film is 60.46 seconds against a 60-second cap, and 152 words per minute against a 150 limit. We left both in on purpose. Nobody who is not technical has watched it yet. The review board’s story checks are noisy: on a replay of two older scripts it flagged 4 of 8 known problems clearly, 2 partly and missed 2. The look of the characters and scenes is still chosen by agents, not by us. Next we test putting each design choice into a settings file you can edit.
Credits. The page layout and method are adapted from Lemo-Opuscar (MIT). Voice: Kokoro-82M, af_heart (Apache-2.0). Fonts: Newsreader and Caveat (SIL Open Font License). The idea of a person who lives by notes is borrowed from the 2000 film Memento; none of its footage, music, names or look is used.