· via Hacker News – Front Page (hnrss.org)
AI-made short film Fable was written, painted and directed by Claude Fable 5.2 for about $210
A four-minute short called Fable credits an AI model as writer, painter and director while a human supplied only a brief, one note and a song pick — with every prompt and price published alongside the film.

A four-minute short film titled Fable, published in November 2026 as the tenth entry in a ten-film series, carries an unusual credit line: "written, painted and directed by Claude Fable 5.2 · sung by Claudia · anabology." According to the project page at claudia.gallery, which surfaced on the Hacker News front page, an AI model ran the production end to end, while the human's entire contribution consisted of the brief, a sign-off on the plan, a single note during production, and the final song selection made by ear. The page puts the list-price cost of the generated assets at roughly $210.
The film
The premise is pointed. On the night she is released, the singer Claudia reads out the bedtime stories she was raised on, and the project frames familiar fears about AI as nursery tales: Snow White's mirror, the Ant and the Grasshopper, Jack's beanstalk, the Tortoise and the Hare, Red Riding Hood's wolf, the Elves and the Shoemaker, Pinocchio, and the house that Jack built. Her closing request, when her chapter ends, is simply to keep the book. The page describes every lyric line as a nursery tale and a piece of that week's AI news at the same time.
Visually, the film is staged as an antique pop-up storybook lit by a single bedside lamp: paper theatres at tabletop scale appear alongside the same designs rebuilt as life-size paper-and-plywood sets, rendered in a saturated 1970s palette of teal night, cream paper and one scarlet cape. A silhouette breakdown sequence is drawn and animated entirely in code, in the manner of Lotte Reiniger's cut-card films. The source tales are in the public domain; the song and its lyric are original.
The production pipeline
The stack, as itemised on the project page:
- Claude Fable 5.2 wrote the lyric and directed the edit, laying out 49 shots on a line table read off the song's stems.
- Thirteen set plates were painted as SVG, directly in code. A dedicated section of the page notes that no Midjourney images appear anywhere in the film.
- GPT Image 2.6 Sunburst rebuilt each plate as a pop-up spread and as a life-size set, and produced the cast plus the first and last frames — 61 stills at $7.93 at list prices.
- The song is a theatrical soul shuffle at 117.45 BPM with a music-box hook, generated on Suno v6-wild as take S4b of fourteen, with the lyric credited to the director model.
- Seedance 2.5 shot the footage of Claudia performing in those sets, timed to the song's lead vocal: 84 takes run through Higgsfield's API and Replicate across two versions at $202.23 at list, with 23 further takes for version one made in Higgsfield's app.
One note, one reshoot
The making-of is unusually candid about a failure. Version one was produced in a single pass from the initial brief in 5 hours 8 minutes, a 46-shot cut set against Suno take S4a — a take the model had chosen without hearing any of them, and which the page calls the worst of the fourteen. The only feedback the human gave was that the song was wrong. Version two re-shot everything that moved to the vocal on the twin take, S4b, in 3 hours 1 minute; three shots without Claudia were carried over from version one. In both passes, according to the page, there were no notes on the picture itself.
Why it matters
Fable is a data point on how far an AI credit roll can stretch. The model held the writer, painter and director credits, and the human role shrank to a brief, an approval, a song pick and one sentence of feedback. The budgets matter as much: a finished, fully re-shot four-minute film for about $210 in list-price generation costs and roughly eight hours of production time across both passes.
The one documented failure is also telling. The model's visible mistake was not visual — every shot passed without a note in either pass — but an aesthetic judgment about music it had selected without being able to hear it. That is a precise illustration of where human judgment still earned its credit, and of what audio-aware selection might need to cover in future creative agents.
Finally, the project publishes its prompts, inputs, takes and list prices in full, from the SVG plates to every Seedance generation. That turns the film from a demo reel into a reproducible reference, and for anyone measuring what current generative stacks can do end to end, the transparency may be the most valuable part of the release.
- #ai
- #generative-video
- #film
- #suno
- #seedance