dashi-animationstyle

dashi-animationstyle (Style Animation) is an open-source Claude Code Skill from Dashi AI Lab (the Chinese channel 大师的AI小灶, GitHub user chuspeeism). You give it a 15 to 60 second Chinese voiceover on a knowledge topic; it transcribes subtitles and writes a storyboard, then Claude draws every shot in code with HTML and Canvas, headless Chrome captures it frame by frame, and ffmpeg assembles a 1920×1080, 30 fps mp4 with the voiceover and subtitles, ready to drop into the video editor CapCut (剪映). No image generation model is involved. It ships with 7 styles: 3D cartoon, dark-background tech, surreal collage, retro paper-cut, retro newspaper, flat flowchart and hand-drawn notes, and the README suggests what each suits, such as flat flowcharts for steps and cause and effect, and hand-drawn notes for breaking down a concept. You can also hand it a reference image to build a new style from, though the README says results aren't guaranteed and it usually takes 2 to 4 rounds of stills; or describe the feel you want in a sentence, and it checks a built-in dictionary of 27 styles and offers stills in 2 to 3 directions to choose from. What sets it apart is that it stops twice for your approval, first on the storyboard and then on 3 still frames, before rendering the whole piece; feedback goes in as "which second, which element, change it to what", and it shows you new stills after each fix.

Who it's for:creators of Chinese-language explainer videos who want animation to go with their voiceover. You'll need Claude Code and the ability to install a few command-line tools yourself; readers who never touch a command line will find it hard going.

Before you start: the author has tested it only on macOS with Apple silicon, not on Windows, where the README says fonts and Chrome paths differ; it has also been tested only with Claude Opus 5.5, and the author hasn't checked how other models do at writing the pages. You need to install Node 22 or later, ffmpeg and Google Chrome yourself; whisper-cpp is optional (for automatic subtitles; a written script works without it), and SKILL.md says the Skill never downloads or installs software for you. Several fonts, such as Lantinghei and Yuanti, are macOS on-demand fonts that you download yourself in Font Book; without them, those fonts fall back to PingFang. On timing, the README says a 20 to 30 second voiceover in an existing style took about 30 minutes in testing, so plan for an hour, and building a new style from a reference takes 1 hour to 1 hour 45 minutes. It suits one 15 to 60 second segment at a time, so pick the parts of a longer video that most need visuals; add music, sound effects and live footage back in your editor, and check whisper's subtitles against your script, since names and English words are the most likely to be wrong. The sample voiceover in the repo, examples/token/口播.wav, is the author's own voice, provided only for trying the Skill; it may not be used in other work and is not covered by the MIT license. The Skill includes no fonts or third-party images; any reference image you supply stays in your own project folder, you are responsible for its rights, and the original image does not appear in the final video. Code and docs are MIT licensed.

Based on the project's public documentation. Not tested by AI Inspiration World.