World Labs (Fei-Fei Li’s new startup) just opened Marble, a multimodal world model that generates persistent 3D worlds from text, images, video and 3D layouts.
The key idea: instead of generating a single object or clip, Marble outputs an entire spatially consistent environment that you can walk through, edit, grow and export as Gaussian splats / meshes / video for use in engines and VR. It also comes with AI-native editing tools and a hybrid 3D editor for blocking out geometry before refining details.
It’s being pitched as a step toward “spatial intelligence” — world models that reason about 3D space for creativity, robotics, and embodied agents. I’d love to hear from people who have tried it or who are building similar systems: what’s still missing for this to be production-ready?
The key idea: instead of generating a single object or clip, Marble outputs an entire spatially consistent environment that you can walk through, edit, grow and export as Gaussian splats / meshes / video for use in engines and VR. It also comes with AI-native editing tools and a hybrid 3D editor for blocking out geometry before refining details.
It’s being pitched as a step toward “spatial intelligence” — world models that reason about 3D space for creativity, robotics, and embodied agents. I’d love to hear from people who have tried it or who are building similar systems: what’s still missing for this to be production-ready?