Graphic tutorial: make your first film with art animation skills, from loading Skill to rendering MP4图文教程：用艺术动画技能做第一支片，从装 Skill 到渲染出 MP4

First install uv (Python package manager), which is the entry point for this skill run script; macOS/Linux is installed with an official line of scripts, and Windows is installed with PowerShell.

After installing ffmpeg, it is required for rendering and audio synthesis; after installing and executing ffmpeg -version, confirm that the publication number can be output.

Install Playwright's Chromium kernel before first use: uv run --with playwright playwright install chromium. This step has a large number of downloads, patience, etc.

Lastly, install the skill ontology: execute npx skills add alchaincyf/huashu-art-motion in the terminal, which will be installed in your agent skills directory; you can also decompress the source code package downloaded from this site and put it into the skills directory.

After loading the Claude Code or Codex, ask "What can this skill do?" If it can clearly explain 35 styles and 9 grammars according to SKILL.md, it means that the loading is successful.

Enter the skills directory and execute this command to render a frame of Van Gogh style: uv run --with playwright python scripts/engine/render.py --solo 09_postimp --stills 0.3 --out trial rendering.

If there is an image file in the directory, it means that the engine, font, and browser kernel are all connected; this step can expose 90% of the environmental problems in advance.

If you get an error saying that you can't find the browser, most likely the playwright install chromium in the previous step is not finished, just run it again.

If Windows users encounter concurrent loading failures, they can check the repair suggestions for Windows concurrent loading in the repository Issues. The author has merged the relevant PR.

Open the references/style recipe/, where each card corresponds to an artistic style (cave frescoes, ancient Egyptian, Van Gogh, Klimt, Bauhaus, Kirby comics, 8-bit, steam waves, Shinkai Cheng, etc.), with parameters, motif actions, signature transitions and current short boards written on the card.

The corresponding executable scene code is scripts/engine/scenes/, and the file name and style correspond one by one; first use the --solo scene name to output a static frame, confirm that the composition is satisfactory, and then do the animation.

Important reminder: The author specifically explained that the 35-sheet overview map uses the same composition, which is the mechanism of "artistic history" (style change, scene unchanged), not the standard composition of these styles. New subjects should be reconfigured for the subject matter, and do not move the room over.

The suggestion process for you is "one frame first": first draw the right moment that best represents this piece, and then let it move, which is much less labor intensive than doing the whole piece from the beginning.

In addition to artistic style, there are 8 narrative grammars in the skill: Kurzgesagt (flat without strokes, scale travel), Vox (newspaper clippings, red lines, highlighter), whiteboard (pen tip unveiling), 3Blue1Brown (one object morphs into the next), storytime (close-up of the reaction, laughing point pause), dynamic text (main words smashed in), launch UI (spot bottom, furry glass card, big numbers), financial charts (first axis, second data, one thing only).

Each syntax is accompanied by a demonstration piece and an example spec. By changing the copy and parameters according to JSON under scripts/engine/examples/, you can output fragments that are accurate to the frame, and support landscape, portrait, and transparent bottoms.

Run an example: uv run --with playwright python scripts/engine/render.py --spec scripts/engine/examples/t3_finance_chart.json --out finance chart.mp4.

If you have an oral broadcast, you can take → the main line of "lens table style → world canvas plus lens → output picture track" and align the animation to your commentary rhythm.

After the rendering is completed, run the acceptance script qa.py, which will measure the stability, efficiency, motion, fluency, and text frame clues of the film, turning "good-looking" into a comparable number.

Then send an agent who is not involved in the production according to the document only as a pick-and-pick question - this step is to bypass the self-identification bias of the creator's perspective.

Finally, use the delivery script: deliver.py will run the reverse PPT access control first, and the red light will not come out directly if there is almost no quick action in the shot; after that, it will be copied into the film and the delivery instructions will be written automatically, and the facts in the instructions will be written by the script.

It is recommended to export a low-code-rate preview for sharing and a high-quality master for retention. If you want to score, there is also a pure code synthesis template in the skill, using BPM grid card rhythm.