Text-to-3D Generation

Text-to-3D generation turns language into 3D objects, characters, interactions, or entire scenes for uses such as design, animation, and virtual worlds. Many methods optimize a 3D representation, such as a mesh, NeRF, or Gaussian splat, using pretrained 2D diffusion models; others use multi-view generation, native 3D diffusion or autoregressive models, and procedural code guided by language models. Work increasingly adds controls for parts, layouts, materials, and physical interactions, while scene-generation systems use iterative visual feedback to correct spatial errors. A central challenge is making outputs both prompt-faithful and geometrically coherent across views, and evaluation remains sensitive to how quality and alignment are measured.