Tendem
Mini toolsDreamFusion
By Google
Instead of training on 3D datasets, DreamFusion “asks” a pretrained text-to-image model what the object should look like from many angles and uses that feedback to refine a neural field until renderings match the prompt. The final radiance field can be meshed and textured for use in engines or design tools, capturing shape and appearance with surprising fidelity from words alone. It’s particularly useful for quickly exploring character and prop ideas, generating turntables, and seeding asset pipelines for games, AR/VR, and product visualization—no paired 3D supervision required.
Video
Gen 1
Overview
DreamFusion is a text-to-3D method that optimizes a neural 3D scene from a text prompt by distilling guidance from a powerful text-to-image model, producing view-consistent assets you can convert to meshes and textures.
About Google
At Google, we think that AI can meaningfully improve people's lives and that the biggest impact will come when everyone can access it.
View Company Profile
Last updated: October 8, 2025
