A new open-source project on GitHub shows how code can turn still paintings into moving scenes. The repository, huashu-art-motion, describes itself as an animation skill for AI agents. It combines 35 art styles with nine explainer-video grammars, and its maintainer, who publishes under the handle alchaincyf, has made the full toolkit available to anyone who wants to build similar films. The repository page, dated October 11, 2026, presents the project as a working toolkit rather than a showcase.
What the project does
According to the repository, the skill draws each scene in code instead of relying on pre-rendered footage. Its 35 art-style recipes range from cave painting and Egyptian murals to Van Gogh, Klimt, Bauhaus, 8-bit and vaporwave. Each recipe includes a runnable Canvas scene, a style renderer and a signature transition.
The second half of the toolkit focuses on explainer videos. GitHub lists eight grammars that draw on the techniques of well-known channels and formats, including Kurzgesagt, Vox, whiteboard, storytime, kinetic type, 3Blue1Brown, keynote UI and finance charts. Each comes with a demo film and parameterized clips that users drive with a JSON spec. A ninth grammar, a presenter-led finance explainer, ships only as a reference implementation and requires users to supply their own character assets.
Long-scroll and analysis tools
Two further components extend the workflow. A long-scroll skeleton lets a character walk from left to right, crossing a boundary where both the world and the painting style change. The repository includes three sample worlds: an Egyptian mural, Monet's Japanese Bridge and an 8-bit scene.
A breakdown script maps a reference animation into cuts, a beat grid and motion heatmaps, so creators can study its timing before building their own version. A QA script then measures stability, cost per frame, motion and smoothness. Before a film is delivered, a delivery script checks it against an anti-slideshow gate, which flags films that contain almost no fast movement within individual shots.
Showcase films
The project's showcase material centres on two flagship pieces. The first is a 65-second film in which a pixel version of the author becomes the hero of a full platform-game level. Bricks, pipes and level movement are coded, while the character is composed from generated frames.
The second, titled Clawd: A Night at the Museum, runs 98 seconds across 16 scenes with no narration. The soundtrack is also synthesized in code. Clawd, the pixel mascot of Anthropic's Claude Code, sneaks through the Metropolitan Museum of Art at night, crosses Central Park to the Natural History Museum and returns before dawn to pose as an exhibit. The five paintings featured, by artists including Van Gogh, Seurat, Vermeer and Rembrandt, are public-domain works from The Met's open-access collection, and each is redrawn in code in its painter's brushwork.
Setup and requirements
Installation runs through a single command, npx skills add alchaincyf/huashu-art-motion. The project requires uv, ffmpeg and Playwright Chromium. The repository's instructions say first-time users should run uv run --with playwright playwright install chromium to set up the browser. The skill's documentation is written in Chinese, which is worth knowing before you start.
Basic code-based animation does not need voice or image accounts. Optional tools for speech and image generation are available, but the repository says they are opt-in and must be chosen by the user.
Compatibility and limits
- Installation and reinstallation have passed in three environments.
- Online regression testing passed in Codex.
- Online regression testing for Claude Code and Kimi was blocked by test account conditions and remains to be verified.
- The ninth explainer grammar and the full-narration examples are code snapshots, not ready-to-render projects.
Licensing and credits
Code and documentation are released under the MIT licence, but the repository lists several exceptions. Bundled stroke data keeps the Arphic Public License, bundled fonts retain their SIL Open Font License terms, and the Huashu character artwork is limited to demonstration use. The Clawd films are an unofficial fan demo, unaffiliated with Anthropic, and are not covered by the MIT licence.
The project credits Tak (@cherry_mx_reds), whose 15-second Art History Speedrun inspired it. The author says that after seeing the original in early October 2026, they asked Claude to recreate it. The repository states that the scene layouts and the girl-and-cat premise follow the original, but every frame is redrawn in code, and it contains no frames, screenshots or audio from the source. The repository also thanks several contributors for fixing a concurrent loading problem in the rendering and QA local services on Windows.
What comes next
For animators, educators and hobbyists, the appeal is a reusable, scriptable pipeline that treats animation as code. Readers should weigh the limits, though. Validation on some platforms is still pending, the documentation is in Chinese, and the character artwork carries restrictions that make commercial reuse unlikely without replacing assets. The project's real test will be whether other creators build on its recipes and report results across more environments.
Related: Reduce Claude Code Token Costs With optimAIzr CLI · Anthropic's OSS Scanner Automates Security Checks for Open Source
Comments (0)