img2threejs turns one photo into a 3D model written entirely in code

One photo in, a spinning 3D model out — and the "model" is just TypeScript you can read.

img2threejs is an open-source skill for Claude Code (Codex and OpenCode work too) that rebuilds the object in a reference image as a procedural Three.js model: primitives, shaders, and generated geometry instead of a scanned mesh file. Per the README, it runs eight self-correcting passes — blockout, structural, form, material, surface, lighting, interaction, optimization — each vision-reviewed against your photo, with quality gates that block generation until the identity-defining details are specced. The repo is at 8.4k stars under Apache 2.0.

Why it matters: the output is code, not a heavy .glb download. Code-only models load fast in any browser, animate cleanly, and you can tweak a handlebar or a blade tint by editing a line. Think of it like commissioning a sculptor who works only in Lego — except the Lego is TypeScript.

Try it: you need Python 3.10+ (standard library only) and one clear photo of the object. Install is a single clone into your skills folder:

git clone https://github.com/img2threejs/img2threejs.git ~/.claude/skills/img2threejs

Then in Claude Code, attach the photo and run:

/img2threejs Rebuild this object as a Three.js model, keep the proportions, angles, and colours.

The pipeline validates the image, writes a spec, generates the passes, and shows side-by-side comparisons as it goes. The live gallery has 10+ finished demos — CS2 knives, a BMX bike, earbuds — all running in your browser right now.

Photogrammetry needs a camera rig and a mesh pipeline. This needs a screenshot and a slash command.