img2threejs

img2threejs takes a reference image and rebuilds the object in it as procedural Three.js code, using primitives and shaders instead of a downloaded mesh. It is an agent skill, not a library or a CLI: you install it into Claude Code, Codex, or OpenCode and the agent runs the pipeline. MIT licensed.

The design is the interesting part. Deterministic Python scripts, standard library only with no external dependencies, handle validation, quality gates, and code generation, while the model's tokens go only to the visual judgment steps: look at the image, assess it, review the render. That split is why it stays token-efficient, and it is a pattern more agent skills should copy.

Web developers who need 3D objects and would rather not model them or license assets are the audience, mostly solo and small teams. Subjects get classified as object, character, or hybrid, and characters route through a separate anatomy-aware track rather than being refused outright. The skill costs nothing beyond the agent subscription you already have.

The catch: it is strongest on hard-surface objects, and the project says so itself. Characters come out as stylized reconstructions, not photoreal likeness, and the projection-first likeness work is still on the roadmap rather than shipped. Credit where it is due, the skill is built to report low confidence and say the fidelity is not reachable from one image instead of confidently handing you something wrong.