Industry reports and Google's official announcements

Google Introduces CAT3D for Rapid 3D Scene Generation with Multi-View Diffusion Models

May 17, 2024, 02:01 AM

Google has introduced CAT3D, a new method that generates high-fidelity 3D scenes from any number of real or generated images in under one minute. This innovative technology uses multi-view diffusion models to create novel views, which are then inputted into NeRF/3DGS systems. CAT3D supports various input types, including text, single images, and multiple images, and can render scenes at 60 frames per second in a browser. The project aims to simplify 3D model creation, reducing the need for hundreds of images typically required for high-quality 3D capture. CAT3D also boasts SOTA few-view results.