huashu-art-motion art animation skills: 35 art styles + 9 narration grammars, let AI use code to really move the picturehuashu-art-motion 艺术动画技能：35 种艺术风格 + 9 种解说语法，让 AI 用代码把画真的动起来

The raw video model is responsible for the "picture", but it is not good at accurately controlling the transition, tempo and typography; in turn, this skill - the main body of the picture is handed over to the raw model for framing, and the "mechanisms" such as bricks, water pipes, level movements, transitions, and camera movements are written in code.

So what it makes is controllable on the timeline: it can be accurate to the frame, it can specify the horizontal and vertical screens, and it can output a transparent bottom, not where the card is drawn.

There is also one thing recorded in the skill: "Which practices have proved effective" (references/07-positive experience.md), each style of pit and short board, which is equivalent to opening the author's pit record together.

In practice, it has anti-PPT access control: there is almost no fast action in the camera, and the red light does not come out, forcing the output to really "move", rather than cutting back and forth on a pile of static pages.

35 artistic style recipe cards, each annotated with parameters, motifs, signature transitions and current short boards, placed in references/style recipe/; 35 corresponding scene codes, in scripts/engine/scenes/.

Explain 9 animation syntax cards, of which 8 come with demonstration films, parameterized fragments and example specs. Feeding a JSON can produce fragments that are accurate to the frame for a period of time.

Reference codes for the entire oral broadcast: mixed style, whiteboard, Vox, and narrator. Some materials need to be prepared (the catalogue README has instructions).

The long roll traverses 1 skeleton (including 3 sections of demonstration and character frame libraries), and 17 painting and animation libraries (brushes, renderers, late-stage, skeletons, lenses, charts, typography, etc.).

The core 12 methodology documents cover disassembly, mechanism, one frame first, pure code rendering, rhythm and soundtrack, characters, long volumes, etc.; optional voice and picture ability access instructions are also available.

qa.py will measure the stability, efficiency, movement, fluency, and text frame clues of the film, and break "good-looking" into measurable indicators.

Another agent who did not participate in the production was only regarded as a film selection problem to avoid the self-confirmation bias of the author's perspective.

The final deliver.py command is delivered: the film is first reversed PPT access control (the whole film is almost red without in-frame quick action, and the film does not come out), after which it is copied into the film and the delivery instructions are automatically written. The facts in the description are written by the script instead of the model.

This process is where the real value of this skill lies - it turns a “craft” into a reproducible assembly line.

The following is the complete source code package of the main branch. After unzipping, the skill directory is obtained, including SKILL.md, references/, scripts/and assets/.

The official recommended installation method is a command: npx skills add alchaincyf/huashu-art-motion, which can be called in Claude Code/Codex after loading into your agents skills directory.

Run dependencies: uv (Python package manager), ffmpeg, Playwright Chromium (first run uv run --with playwright playwright install chromium install browser kernel).

To verify the environment first, you can try rendering a paragraph: uv run --with playwright python scripts/engine/render.py --solo 09_postimp --stills 0.3 --out after entering the skills directory, you can get a frame of the Van Gogh style paragraph. If you want to export directly, use --spec to point to the JSON configuration under examples.

The skill content is mainly in Chinese (with English README paragraph), which is quite friendly to domestic users. There are 35 full-style sample MP4s in the warehouse on the Releases page. You can download them to see the effect before deciding whether to to toss the environment.

This site organizes resources from the GitHub open source repository alchaincyf/huashu-art-motion (mit license), for learning and technical exchanges only. Note that there are exceptions in the license: the Arphic Public License is used for pen-order derivative data, and the respective SIL OFL license is used for built-in fonts. The cartoon image and character frame of Uncle Hua (including the overview map and the same image appearing in the demonstration video) are limited to this skill demonstration, and are not used for other purposes under the mit license. The repository does not contain any frames, screenshots, or audio from the original author Art History Speedrun, so please respect the rights of the original creator as well.