DreamFusion is a cutting-edge tool that uses Imagen to turn captions into 3D scenes. The core innovation, Score Distillation Sampling (SDS), optimises samples in 3D parameter spaces. DreamFusion smooths the transition from text to image to 3D using a 3D scene parameterization similar to NeRFs. The tool optimises geometry and scene appearance using various methods. The Lambertian shading model can illuminate the final 3D models, which have superior normals, surface geometry, and depth.
User objects: 3D artists, game developers, filmmakers, architects, and virtual reality content creators.