What is three.js-object-sculptor-codex-plugin?

Three.js Object Sculptor is a plugin for Codex, an AI coding assistant, that turns a picture of an object into a 3D model built entirely out of code, using a JavaScript library called Three.js that renders 3D graphics in a web browser. Instead of scanning the object or downloading a ready made 3D asset, it guides the AI through a step by step process: check that the image is usable, describe the object in detail, break it down into its shapes and materials, build it from a rough blockout up to fine detail, connect the parts so they can be animated later, then compare a render of the result against the original photo. The plugin is meant for objects that will move or interact in a scene, such as game props, mechanical parts, destructible objects, or even plants and trees, rather than for exact scans. The README is upfront that it is not meant for photogrammetry or for pulling a perfect mesh out of a single photo. Its output is a structured plan called an ObjectSculptSpec that records the object's parts, materials, lighting, and the pivot points needed for later animation, followed by generated code that builds the 3D model. The project includes checks meant to catch a common failure in AI generated 3D models, where the overall shape looks right but small identifying details are missing or wrong. It compares a screenshot of the rendered model to the original reference image using an AI vision reviewer and can block progress if an important feature does not match closely enough. To use it, someone needs Codex with local plugin support and Python 3.10 or newer installed, and it is set up by cloning the plugin into a local plugins folder and registering it in a Codex marketplace configuration file. The helper scripts rely on Python's standard library and do not need a full browser automation tool installed. Two live demos, a tower ship and an autumn tree, show the kind of output the plugin can produce.