img-to-3d

Analyze the reference image for shapes, proportions, and materials, then write an ObjectSculptSpec JSON describing primitives like boxes, spheres, and cylinders. Build the model through blockout, structural, material, and lighting passes, then verify by comparing a rendered screenshot against the original image.

Procedural modeling builds geometry from code-defined primitives and transforms, producing lightweight, editable, animation-ready assets. Photogrammetry reconstructs meshes from many photos, yielding heavy meshes unsuitable for code review. This approach uses only primitives, no depth maps or point clouds.

A single image constrains accuracy to visible surfaces and estimated proportions. The workflow compensates with a visual verification loop that screenshots the rendered scene, compares it to the source image with a vision model, and iterates on layout, proportion, and color differences.

Yes. The output TypeScript factory returns a THREE.Group that can include sockets, colliders, and animation groups defined in the spec. The optimization pass merges geometry and reduces draw calls so the asset stays performant for interactive web scenes.

Avoid it when you need photorealistic surface detail, organic scans, or exact replicas of complex geometry, since primitives approximate rather than reconstruct. Use photogrammetry or mesh reconstruction tools for those cases instead.