One reference image in. A code-only, animation-ready Three.js model out.
Reconstruction by code — not photogrammetry, not mesh extraction, not downloaded art packs. Every model this org produces is a TypeScript factory function you can read, diff, and animate.
| Project | What it is |
|---|---|
| img2threejs | The agent skill itself — a staged, quality-gated sculpting pipeline that turns one reference image into a procedural THREE.Group factory. Runs under Claude Code, Codex, or OpenCode. Python 3.10+ stdlib only, zero dependencies. |
| img2threejs-showcase | The live gallery — every demo is generated code running in your browser. Orbit any model and read the source it was rebuilt from. Contributions welcome via PR. |
- Code, not meshes. Output is diffable TypeScript plus a JSON spec — no multi-megabyte binaries in your repo.
- Animation-ready by construction. Every model ships a runtime hierarchy: pivots, sockets, and colliders, so it can move the moment it exists.
- Detail-first, gate-enforced. A
detailInventoryof identity-defining details (bevels, rivets, engraved linework, wear, gloss) must map to real components before a line of Three.js is generated. - Token-efficient on purpose. Deterministic Python scripts do validation, gating, spec authoring, and comparison-sheet packaging. Model tokens are spent only on visual judgment and code.
- Honest about limits. One image cannot reveal hidden sides. The pipeline reports per-region confidence and says so plainly instead of faking fidelity.
One theme per release, so there's always a single answer to "what is this version about".
| Version | Theme | What it unlocks |
|---|---|---|
| v1.4 | The Weapon Update · shipping soon | 1:1 photorealistic hard-surface assets — strict PBR + texture projection, mechanical reasoning, metal/polymer/wood |
| v1.5 | The Character Update · in progress | Character reconstruction, facial features, rigging-ready topology, blendshape prep |
| v1.6 | The Environment Update | Buildings, rooms, streets, vegetation, terrain-aware and multi-object reconstruction |
| v1.7 | The Game Pipeline Update | Unity / Unreal exporters, Blender bridge, LOD + collision meshes |
| v1.8 | The Animation Update | Auto rigging, skin weights, Mixamo compatibility, facial rig |
| v1.9 | The AI Studio Update | Web UI, batch processing, visual prompt builder, cloud rendering |
| v2.0 | The Procedural World Update | Multi-view reconstruction, procedural cities, semantic world understanding, plugin API |
The arc: assets (v1.4–v1.5) → worlds (v1.6–v1.7) → production, rigging and engine exporters (v1.8–v1.9) → an AI game-asset platform that generates playable worlds from reference images (v2.0).
Full breakdown: ROADMAP.md · roadmap on the site
git clone https://github.com/img2threejs/img2threejs.git ~/.claude/skills/img2threejsThen, in your agent, point at an image:
/img2threejs Rebuild this object as a Three.js model, keep the proportions, angles, and colours.
The pipeline probes the image, writes an assessment and spec, generates the factory pass by pass, and shows you a reference-beside-render sheet at each step until it matches.
New procedural material recipes, gates, host coverage, and showcase demos are all welcome. Start with CONTRIBUTING.md and the roadmap, or open a Submit a showcase demo issue to get feedback on scope before doing the work.
img2threejs is free and open source. If it saved you time, a coffee helps keep it going: Buy Me a Coffee.
Logo, avatar, and banner assets live in brand/
with usage notes.
