Edify 3D is a foundation model published by NVIDIA Research that converts text descriptions or reference images into fully textured, structured 3D assets. The pipeline operates in two main stages: first, a set of multi-view diffusion models generates RGB images and surface normal maps of the target object from multiple camera angles simultaneously; second, a Transformer-based reconstruction model uses those multi-view observations to predict geometry, texture, albedo, and PBR material channels, then extracts a clean quad mesh via isosurface processing. The result is a 3D asset with well-organized mesh topology, automatically generated UV maps, and up to 4K physically based rendering textures, completed in approximately two minutes from input to output. Assets use the OpenUSD format, enabling direct import into NVIDIA Omniverse USD Composer and compatible 3D pipelines.
Edify 3D has been adopted in production workflows through partnerships with Shutterstock and Getty Images, both of which built commercially licensed generative 3D services on the Edify architecture. Game developers use it to rapidly prototype environment props and secondary scene objects. Architects and interior designers apply it to generate furniture and material variants for visualization. VR and XR studios use the model to quickly populate immersive scenes with custom assets. The technology was demonstrated publicly at SIGGRAPH 2024, where NVIDIA researchers used it live to generate dozens of distinct 3D objects—including rocks, cacti, and decorative items—within a few minutes as part of a full scene composition workflow.
Explore this option.
Associated Tags
text to 3D, image to 3D, 3D asset generation, PBR materials, multi-view diffusion, NVIDIA AI, OpenUSD, game asset generation