A DeepSeek Harness skill that turns Chinese copy, style prompts, and audio into coherent code animations (Remotion / Canvas / HTML): it first reads the full text to create semantic storyboards, then defines the visual style, camera angles, transitions, dynamic text, and backgrounds, and finally renders and checks for revisions, aiming to avoid "one sentence per card" slideshow-style animations.
Stars 54
Reasons
- Fixed workflow, no skipping steps: Read the full text → Segment by content and highlight key terms → First ask the author whether to maintain a unified style or change styles per segment → Retrieve and fully read the selected style source → Storyboard shot by shot → Implement; deliver in four tiers based on authorization each time (task card only / static frames only / audio preview / final video), and choose the smaller tier if the scope is unclear.
- The package includes method documentation (modules/), specific recipes (recipes/, such as transitions, camera movements, paper puppets, charts/maps, UI effects), reusable source code (implementations/, including cameras, characters, transitions, Chinese text components, and Remotion suites), task card templates, 7 complete examples, and task card validation and motion logic test scripts.
- Third-party sources are listed item by item in LICENSES/README.md (Renoise, LottieFiles, HeyGen, etc., MIT / Apache-2.0 snippets, with fixed commit hashes); 10-30 library setup, first inclusion in this column.
Usage
- git clone https://github.com/Akillll-DuoHe/fy-code-animation.git，把整个目录复制进 into the DSH skills directory (e.g., ~/.dsh/skills/); a Gitee mirror is available for users in China, and a v1.0.0 archive is also available in Releases.
- For first-time use, run python scripts/doctor.py --root <your style library> for a self-check, then run python scripts/style_library.py build --root <library root> to rebuild the index; Python 3.10+ is required, Node 20+ is required for test scripts, and rendering can use Remotion / Canvas / HTML as preferred, with dependencies provided by your project.
Risks
- The core dependency, the "341 style library," is not distributed with the package, and the README does not specify where to obtain it; without this library, the style retrieval step cannot proceed, and the skill will stop to ask you for the path.
- Not a standard open-source license: The author only allows free use and sharing, and prohibits packaging it into paid courses or paid skills for sale; GitHub cannot identify the license type (NOASSERTION); the README lacks final video samples, so you must create one yourself to see the output quality; the music solution relies on local ACE-Step, which requires self-deployment.