AI Decision System
Replayable, offline-verifiable records for AI-assisted go/no-go decisions: experiment promotions, typed model answers and human-approved actions.
翻譯準備中
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證Scroll-driven websites from one YAML file: generated film, live 3D, data stories, maps. Runs locally, no API keys; Claude can drive it via MCP.
翻譯準備中
取自儲存庫 README
正在載入完整 README。
檢視完整 README ↗| Scenes | Your brand | Workflow |
|---|---|---|
| One camera take through a whole page | Remake any landing page: studio remake <url> | Six art directions from one brief |
| Cinemagraphs, light from dawn to night | Your colours, fonts and logo from a URL or a logo | Storyboard before you render |
| Fly inside your product's 3D model | Every spacing, type and motion value is yours to set | A polish report on every film |
| Miniature worlds, maps, charts, live 3D | Optional experiences on top: 32 styles, from zoetrope to paint on glass | Claude drives it all through MCP |
| Photo parallax from your own pictures | Any brand, from a page, a logo or a brand.yaml | Remake a live landing page: studio remake <url> |
The heavy scenes (generated film and images) are made with open models on your own GPU (Z-Image, Qwen-Image-Edit, LTX-2.3, Wan 2.2, SeedVR2, Blender); photo parallax, kinetic type, maps, data, code, live 3D and every style need no GPU. The output is plain HTML, CSS and JS you can host anywhere. The examples show what each scene does; they are not templates. Bring your own photos, footage, 3D models, logo and copy, or a brief.
git clone https://github.com/Okohedeki/scroll-studio && cd scroll-studio
pip install -e .
studio doctor # what this machine can build
studio build examples/cellwright-bio # live 3D: needs nothing but Python
studio preview cellwright-bio # http://localhost:5173
studio ui # the Studio app: http://localhost:5180
No GPU? Try it on a laptop: studio new my-site --laptop makes a site that builds without ComfyUI or
Blender, and studio remake <url> --no-gpu turns a live page into one.
| What you build | What it needs |
|---|---|
| Live 3D, charts, maps, type, vector art, splat scans, miniature worlds, code flights, all 32 styles | Python and a browser |
| Photo parallax, product sequences, inside the product | Runs on the CPU; Blender and a GPU make it fast |
| Generated film, the one-take page, cinemagraphs, light as the scroll, generated images | ComfyUI and a 24 GB GPU (RTX 4090 class) |
Full setup (GPU, Blender, ComfyUI and the models) is in docs/INSTALL.md.
Open the repo in Claude Code and ask for a site. The scroll-studio MCP server (.mcp.json) gives Claude every
engine tool, and the scroll-site skill gives it the workflow. Or paste this into any coding agent:
Install Scroll Studio on this machine and build one of its examples, following
https://github.com/Okohedeki/scroll-studio/blob/main/docs/INSTALL.md
studio new my-site # a starter site.yaml
studio remake https://example.com/ # or start from a live landing page, told as a scroll site
studio directions my-site # six art directions; --pick one
studio storyboard my-site --edit # sketches, stills and an animatic before any video render
studio build my-site # cached stage by stage: change one thing, rebuild one thing
studio inspect my-site # pops, softness, halos and drift in every film
studio publish my-site --out site # one static folder for GitHub Pages or any host
The whole workflow is in the guide.
| Need | Start here |
|---|---|
| Setup | Install · What each machine can build |
| What it can make | Scenes · Styles · Examples · Showcase |
| Reference | site.yaml, CLI and MCP reference, generated from the engine |
| Making a site | Guide: direction boards, storyboards, client copy decks, trailers, Claude |
| Why it exists | Local only, 10 to 30 times fewer tokens |
| How it works | Engine · Style research · Roadmap |
| Contributing | How to contribute · Rules per area · · |
Free software under the GNU AGPL-3.0 with additional terms (NOTICE.md): derivatives stay open source and keep the attribution word for word, and sites you build are yours to license however you like. Model weights keep their own licences. Details in docs/licence.md; AI agents, see AGENTS.md.
| The project | Governance · Maintainers · Compatibility · Design proposals · Security · Threat model |
| Releases | What's new · Changelog · 1.5 milestone |