3D Generation
Creating 3D models - meshes, point clouds, or implicit representations - from text, images, or other inputs.
Think of it like
Sculpting: a 2D sketch or description -> a 3D form.
Example
"A wooden chair, modern design" - generates a textured 3D model. Or: a 2D image -> 3D reconstruction.
How it actually works
Text-to-3D: use pre-trained 2D diffusion (e.g., Stable Diffusion) to score candidate 3D models (NeRF, mesh), optimize via SDS (score distillation sampling). Single-image 3D: depth estimation + shape completion. Challenges: slow (many iterations), unrealistic fine details, lighting ambiguity. Implicit representations (NeRF) vs. explicit (meshes). Recent: diffusion models directly in 3D space (e.g., 3D diffusion models).
For product teams
Accelerates game art, architecture visualization, e-commerce product renders.
For engineers
Methods: NeRF + SDS loss, diffusion-based, flow-based. Representations: mesh, voxels, point cloud, SDF, implicit neural fields.
Read anything AI without the jargon
Look up any term in plain English, or save terms as you read with the free Chrome extension.
Open DecoderAdd to Chrome