GitHub radar
img2threejs: Turn Any Photo into a 3D Three.js Model via AI Agent
A Claude Code / Codex / OpenCode skill that takes a reference photo and outputs animation-ready Three.js TypeScript code — no 3D software, no mesh downloads, no designer.
img2threejs is an AI agent skill for Claude Code, Codex and OpenCode, published in mid-July 2026 and reaching 12 400 GitHub stars in five weeks. It takes one reference image of an object and produces a TypeScript THREE.Group factory that rebuilds the object from primitives, procedural shaders and generated geometry — animation-ready, with pivots, sockets and colliders baked in. The pipeline runs eight stages (blockout → structural → form → material → surface → lighting → interaction → optimization), each gated by a vision review before advancing. Helper scripts are pure Python 3.10+ standard library — no pip install, no extra dependencies. The model spends tokens only on visual judgment and code; JSON validation, comparison sheets and manifest generation all run in deterministic scripts. Supports general hard-surface objects, characters (with facial landmark fitting and per-region confidence reporting), and CS2-class weapons (dedicated component contracts and family-specific review gates).
Why a vibe-coder should care
If you need a 3D element on your site — a product visualization, a character, a scene prop — img2threejs lets you get it from a photo without 3D software, a designer, or downloading heavy mesh files. The result is diffable TypeScript code that lives in your repo and version-controls like any other file.
Runs on a regular laptop — only Python 3.10+ and Claude Code (or Codex, or OpenCode) needed.
Open on GitHub▌ More finds