The Brothers Grimm's "Cinderella" (Aschenputtel) — KHM 21, twenty-first of a project to film all 240 Grimm tales, one by one, in canonical order. 100% AI-made on one home GPU (an RTX 3090), $0 marginal cost: nothing was sent to a cloud service and nothing was paid for.
This is the Grimms' Cinderella, not the pantomime: no fairy godmother, no pumpkin, no glass. A dying mother makes one promise; a hazel branch planted on her grave and watered with tears grows into a tree with a white bird in it; the tree throws down silver and then gold; a slipper of gold sticks in pitch on a staircase; two step-sisters cut off a toe and a heel to force it to fit, and two pigeons on the hazel call the blood out loud until the true bride rides past — and at the wedding the same pigeons take the sisters' eyes.
The text is Margaret Hunt's 1884 translation (public domain) with whole-sentence cuts only: every kept sentence is verbatim — the blessing, the lentils, "The good into the pot, the bad into the crop," "Shiver and quiver, little tree," both knife lines, and all three verses of "Turn and peep."
The score is a chorale with variations: one hymn phrase — the mother's blessing, stated at the deathbed — and every other music in the film derived from it. The hazel tree is the chorale arpeggiated in harp; the wish-verse Cinderella sings is the chorale's own melody; the festival waltz walks on its bass; and the pigeons' verdict is sung by flute and oboe in close thirds over a ride whose waltz has lost its downbeat — a limping dance for a maimed foot, with the knife as a col legno tick. The birds' pecking music is sweet glockenspiel mercy in the kitchen and returns fortissimo at the church door as the sentence. When the shoe fits, the whole film turns to D major and the blessing comes back tutti; two violin harmonics land on her shoulders with the doves and stay to the last chord. Sonatina Symphonic Orchestra through sfizz; a real cadence, no fade-out.
Eight voices from seven bases, and the one shared base is the tale's reading: the dying mother and the two judging pigeons are the same voice — she watches from heaven and speaks through the birds. The diegetic layer is synthesized end to end: the hearth, lentils into ash, all the birds beneath the sky whirring in at a kitchen window, dresses falling out of a tree, an axe on a pigeon-house, a slipper in pitch, hooves out and back, wash-water, wedding bells, and two doves striking at the church door.
Each of the 13 shots opens on a still from FLUX.1-dev at native resolution, judged by a local vision model (best of 3-4 candidates, forced side-by-side discrimination with defect flags), conformed to locked character sheets with Qwen-Image-Edit-2511 (from this film on through an 8-step distill — a speed change A/B'd against the previous film before adoption), and continued into real motion by Wan 2.2 — 123 generations. The consistency layer judges every conformed key frame for character identity (clothing against the scene's own plate, the face against the sheet — Cinderella wears five costumes and one face) and restores chained animation frames before they seed the next clip. The narration is the clock: each line's measured duration sets every cut and every bar.
Build time: 877h 21m, unattended.