The Brothers Grimm's "The Riddle" (Das Rätsel) — KHM 22, twenty-second 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.
A King's son rides out with one faithful servant and sleeps a night in a witch's house against her step-daughter's warning. At parting the witch brings a glass; it breaks, the poison kills the servant's horse, a raven eats the horse, and twelve murderers at a forest inn eat the raven in their soup and fall dead at the table. From that chain the prince makes a riddle — "One slew none, and yet slew twelve" — and lays it before a proud princess who has already sent nine suitors to the block. Three nights of eavesdropping, two mantles torn away, a bundle of rods, and a misty-grey mantle held fast in the dark decide it.
The text is Margaret Hunt's 1884 translation (public domain), UNABRIDGED: every sentence verbatim and in order.
The score is a riddle-canon for chamber orchestra. One unresolved tritone in the horn is the RIDDLE and hangs over the film until the last twenty seconds, when it lifts to the fifth of the wedding key. The POISON is a single chromatic figure handed on exactly as the poison is: from the glass (glockenspiel and a harp glissando shattering downward) to the horse (low strings falling to the bottom E, the pulse stops) to the raven (oboe, with a clarinet croak) to the twelve (a string canon cut off entry by entry with a bass drum until nothing is left). The road is a passacaglia ground in E minor that slips a tone down into D minor when it reaches the princess's town and becomes her sarabande; the three nights are a nocturne of low harp over one held harmonic; the misty-grey mantle is a chord of string harmonics, torn away in a single pizzicato; and at the verdict the sarabande turns to D major embroidered with glockenspiel and harp. Sonatina Symphonic Orchestra through sfizz; a real cadence, no fade-out.
Seven voices on seven bases, directed line by line: the prince gives his two answers awake while pretending to dream; the witch is a growl trying to be friendly; the princess is proud by day and a whisper in the dark; the twelve judges are one deep voice in stone. The diegetic layer is synthesized end to end: hooves and forest, the fire blown up under the pot, the glass breaking and the horse falling, the raven's wings, twelve at table and the spoons, pages turning, the rods, the mantle torn, and the court's bells.
This is the first film built under a new rule: the art must match the story. The narration is cut into 45 beats, one per narrated event, and each beat has its own picture whose prompt names the face of that moment. 36 of the 45 shots are stills — a FLUX.1-dev plate at native resolution, judged by a local vision model against the beat's intent, conformed to locked character sheets with Qwen-Image-Edit-2511, judged again for identity, hands and head-count, upscaled and given a slow Ken Burns move. Only nine shots — the riders setting out, the girl with the lantern, the glass and the horse, the servant's run, the raven, the twelve falling, the ride from the inn, the rods, the mantle torn — are continued into motion by Wan 2.2 (24 generations). The narration is the clock: each line's measured duration sets every cut and every bar.
Build time: 13h 18m, unattended.