Google DreamFusion

Google’s version of generative AI for 3D models requires no training on 3D model data, but they also generate 3D models a little differently than the other platforms and are not the go-to tool for game development.

The system uses 2D images of an object generated by the Imagen text-to-image diffusion model to understand different perspectives of the model it is trying to generate. This process is called Score Distillation Sampling (SDS) by Google Engineers. SDS creates the basic appearance and DreamFusion optimizes the assets to fill in the model such as adding regularizers and improving geometry. Once processed, these models have high quality normals and can be lit like regular 3D models

Last updated