GET3D

GET3D generates production-ready 3D textured meshes from 2D image collections. Developed by NVIDIA's Toronto AI Lab and presented at NeurIPS 2022, this generative model combines explicit mesh geometry with physically realistic textures in a single end-to-end process.

Highlights

• Generates complete 3D meshes with separate geometry and material properties
• Handles diverse object categories from vehicles to characters to buildings
• Supports text-conditioned generation by combining with CLIP embeddings
• Creates view-dependent lighting effects without manual material authoring
• Produces interpolable latent spaces for smooth shape transitions

The model is open source and available on GitHub for research and prototyping. Commercial deployment and licensing requires formal registration through NVIDIA. GET3D is ideal for 3D game developers, digital asset creators, VFX studios, and research teams needing to automate 3D asset generation from 2D training data.