seedance-skill
alchaincyf/seedance-skill
seedance-skill is an agent skill project by alchaincyf on GitHub: Clay-render-driven Seedance workbench: renders clay reference videos from scene.json with geometry checks, can reverse-engineer a clay model from a photo, and writes prompts from the clay geometry. It covers storyboard and prompts, with prompts for Seedance / Jimeng, and its README says it works with Claude Code, Codex, Cursor and OpenClaw. 40 stars, last updated 2026-10-13, MIT license.
Clay rendering requires Node, ffmpeg and hyperframes
Editor's notes
Up front it asks only for length, aspect ratio and whether to cut, lays out its other inferred parameters in a table for you to veto, then shows a 3–5 line shot list to edit. The clay-render path: write scene.json, run probe.mjs for pre-render geometry checks (subject out of frame, subject too small for the model to keep, camera inside a wall), render the clay video with hyperframes and pull frames to check it, then make_prompt.mjs writes the prompt from the measured shot size and camera distance. Camera moves are solved in solver.mjs, shared by the renderer and the checks. It needs Node 18+ and ffmpeg; photo reconstruction also needs Python with opencv-python.
- Type
- Storyboards · visual assets · video prompts
- Stages
- Storyboard · Prompts
- Video models
- Seedance / Jimeng
- Agents
- Claude Code · Codex · Cursor · OpenClaw
- Format
- Agent skill
- Language
- Chinese
- GitHub stars
- ★ 40
- Last updated
- 2026-10-13
- License
- MIT
- SKILL.md files
- 1
- Checked by hand
- 2026-10-13
In the author's words
Seedance.skill: a clay-render-driven AI video workbench. The clay reference video and the prompt come from the same scene.json, geometry is checked before rendering, and it can reconstruct a clay model from a single photo. Control AI video with clay renders, and a prompt that can't drift from them.
Translated from the Chinese original.
Skills in this repo (1)
Names and descriptions come from each SKILL.md frontmatter; Chinese descriptions are shown in our English translation.
- huashu-seedance
End-to-end Seedance 2.5 video generation. It first clarifies the request into parameters (duration / aspect ratio / camera type / follow mode / whether to cut), then uses the scene.json engine to render a clay reference video (with geometry checks before rendering), can also reconstruct a clay model from a photo the user provides, and finally writes a prompt strictly aligned with the clay render according to the official guidelines.
Install
npx skills add alchaincyf/seedance-skill
# pick agents (claude-code / codex / cursor / opencode / gemini-cli / openclaw)
npx skills add alchaincyf/seedance-skill -a claude-code -a codexDefault commands use the skills CLI and were not verified one by one; follow the repo README. Read SKILL.md before installing. Install guide →
Similar projects
- visual-skills★ 521
AI film director: dramaturgy rules such as Murch's editing principles, blocking and montage, plus exact Seedance prompt syntax
Toolkit from enterprise AI video projects: Seedance/Kling prompts, storyboards, model selection and debugging, with 600+ prompts and methodology write-ups
- DirectorSKILL★ 234
AI filmmaking: shot lists, keyframe and video prompts, with 20 built-in director styles such as Spielberg
- h3-storyboard-skill★ 178
MiniMax H3 shot breakdown and character performance design
Shot design for Seedance 2.5