Procedural 3D Story — workflow skill for 3D scenes
For AI agents: find the right skill in one call —
curl -s "https://skills.varg.ai/api/search?q=<what the person wants>"returns the matching use case and the best skills with install commands, as markdown (Russian works too). Full index: https://skills.varg.ai/llms.txt. Every page has a markdown twin: add.mdto its path. Before generating anything, show the person the options on the use-case page.
Turn a myth or book into a seeded, multi-chapter procedural three.js film in one scene.html, audited frame by frame.
- Type: Workflow · 3D scenes
- Slug:
karpathy-3d-story - Status: beta
- Publisher: chencheng (云谦) (community)
- License: MIT
- Source: https://github.com/sorrycc/skills/tree/master/skills/karpathy-3d-story
- Works in: Claude Code
- Requires: Google Chrome (shot.sh uses the macOS app path), Python 3 with Pillow; ffmpeg and ffprobe for audio, Node.js (audio_sweep.js), ElevenLabs API key (ELEVENLABS_API_KEY) for the optional soundtrack
- Cost: Free — renders locally; the optional narration, ambience and SFX are billed by ElevenLabs
- Adoption: 32 GitHub stars (GitHub, 2026-10-26)
What it does
Builds a multi-chapter animated 3D story (default 5 chapters of about 44 s) as one self-contained scene.html. It contains procedural three.js sets, one biped rig dressed in costumes, an FX library, a director with camera splines and captions, and optional ElevenLabs narration, ambience and SFX. Every frame is a pure function of story time with a seeded RNG, so ?chapter=N&t=X&freeze=1 reproduces any frame. shot.sh renders that frame synchronously in headless Chrome with preserveDrawingBuffer and a toDataURL readback shim. The skill requires at least three screenshot audit rounds over every camera keyframe, and a bundled Journey to the West build is the reference implementation (the idea comes from a Karpathy post). Caveats: the output is a playable HTML film, not an MP4, so capture it with your own frame loop. It also downloads three r160's old UMD build, hard-codes the macOS Chrome path, and suggests pkill -9 -f "Google Chrome", which closes your own browser too.
Use when
- A 3D story or procedural three.js film from source material
- An animated diorama of a book, legend or myth
- Adding a chapter to a story built with this skill
Not for
- Direct MP4 delivery: it ships an HTML film, so capture it through HyperFrames, Remotion or a headless frame loop
Install
- Claude Code:
npx skills add sorrycc/skills
Models it drives: eleven_multilingual_v2
Example output
https://skills.varg.ai/previews/karpathy-3d-story.mp4 — The creator's Journey to the West example for /karpathy-3d-story (5 chapters, 3.5 min of procedural three.js): the camera rises over Flower-Fruit Mountain as the stone egg glows and a beam of light breaks out. Cut 17-25 s of 218 s. (credit: 陈成 chencheng (sorrycc, @chenchengpro), https://x.com/chenchengpro/status/2085274449121788324)
Used for
- AI music video: A music video for your song: concept, a consistent lead character, shots cut on the beat, lip-sync on vocals.
More 3D scenes skills
- Three.js Skills: 13 Three.js skills: scenes, geometry, materials, lighting, GLSL and TSL shaders, animation, post-processing, R3F.
- Three.js Scene Composition: Plan, build and capture Three.js scenes for video: absolute-time frames, seeded motion, color-safe glTF, render QA.
- R3F Skills: 11 React Three Fiber skills checked against Fiber 9.7 and three r185: Canvas, demand loops, loaders, shaders, post.
- Three.js Architecture Effects: Self-assembling 3D buildings in three.js: bricks, timber, eaves and PBR, driven by one seekable progress value.
Run the generation on varg
varg is one API key for the models these skills drive — Seedance 2.5, Kling 4, Veo 3.1, Wan 3, Nano Banana 2.1, GPT Image 2.5, ElevenLabs v4 and 200+ more — billed in credits, with a cost estimate before anything runs.
- Claude Code:
claude plugin marketplace add vargHQ/skills && claude plugin install varg@varg(skill + hosted MCP) - Codex, Cursor, Gemini CLI, OpenClaw:
npx skills add vargHQ/skills --skill varg-ai -a <agent> -y - Claude.ai, ChatGPT, any MCP client: connect
https://mcp.varg.ai/mcp - API key: https://varg.ai/dashboard/api · Model specs: https://varg.ai/models/llms.txt