Huashu Art Motion is an installable agent skill that makes a coding agent write and render animation code in a named art style, such as Van Gogh impasto or Mario-era pixel art. It is aimed at people who already drive a coding agent and want short animated video out of it instead of still images. The repository holds the instructions, parameterised clip templates and reference code the agent reads; it is JavaScript, MIT licensed, and it is not an app with a timeline you click.
What it does
You hand the skill one art style and a subject, the agent writes the motion code that draws it, and the result is rendered out as video.
The counts come from the project itself: 9 narration grammars (ready-made ways to structure spoken commentary over a film), 8 parameterised clip types, and reference code for one complete voiced piece. The style count is inconsistent in the materials: the GitHub description says 35 art styles, while the README header says 36. Either way that is the author's own count, not an independent audit.
One piece is worth singling out. The skill includes a gate that rejects motion which is really a slideshow, so only continuous movement gets rendered and shipped. That is the author's description of their own pipeline, and the showcase is what backs it up.
The showcase is two films. The first puts the author into a pixel platformer for 65 seconds: bricks, pipes, level scrolling and transitions are produced in code, and the character comes from generated frames composited in. In one beat, grabbing a power-up means first picking Claude over OpenAI. The second is a 98-second museum-at-night film starring Clawd, the Claude Code pixel mascot, across 16 scenes with no voiceover and a soundtrack the project says is also synthesised in code.
How it works
The interesting detail is how a style becomes motion rather than a filter. In the Van Gogh scene, the brush direction is taken from the structure tensor of the original painting ("Wheat Field with Cypresses"), and the character is repainted as thick impasto on top of that field. Style is encoded as rules a program can follow, which is why a code-writing agent can apply it to new motion.
The dependency list tells you the shape of the render pipeline: a Chromium to draw frames, ffmpeg to encode them into video, and uv to run Python tooling. The copy of the README saved here is cut off before the pipeline is documented in full, so treat that as the outline rather than the exact division of labour.
Getting started
Installation is one command, run where your coding agent can see it:
npx skills add alchaincyf/huashu-art-motion
What it needs on your machine, per the project: a coding agent, uv, ffmpeg and Chromium. Beyond that the project doesn't list hardware requirements, so there is no stated operating system, chip, memory or GPU floor to go by.
On money: the skill itself is free and MIT licensed, and nothing in it requires a paid service of its own. The intelligence is not free, though. It runs on whatever model your coding agent uses, which for most readers means a paid subscription or an API key; the materials don't name a model or a price, and token spend on a multi-scene film is the obvious variable.
If you just want to see the output, the finished sample films are published as downloads under the repository's Releases, and the showcase GIFs sit in the repository itself. The main README is written in Simplified Chinese, with an English translation at README.en.md.
When to use it / when not
The main catch is the toolchain. This is not a one-click thing: you need a working coding agent plus three command-line dependencies installed before a single frame renders. The second catch is age. The repository was created on 2026-10-06 and last pushed on 2026-10-16, so the roughly 5,500 stars reflect attention, not a long track record.
Use it instead of asking Claude directly for an animation script when you want a consistent house style and a check that stops the output degrading into a slide deck. Don't use it if you have no coding agent, won't install command-line tools, or expect a visual timeline of the kind After Effects or Premiere gives you.
Alternatives
The closest named relative is in the materials themselves: in October 2026, the motion-design skill inside nexu-io's Open Design workspace listed Huashu Art Motion as one of its design sources and shipped an adapted copy, keeping the MIT licence and upstream attribution. If you already work in Open Design, you may have a version of this in reach.
Otherwise the alternatives are the obvious ones: write the animation yourself in a code-based video framework, prompt your agent from scratch each time without a skill, or stay in a conventional editor.
Take this seriously if you already have a coding agent set up and you make short video: it is a concrete answer to the question of how to get movement, not just pictures, out of a model. The showcase films are specific enough to judge for yourself before you install anything, which matters for a project this new. If you are looking for a standalone animation app, or you want something with a settled release history, this is not that yet.