DeepMind has released an experimental prototype that allows users to generate and explore interactive environments in real-time. Powered by the Genie 3 model, the tool enables users to create playable worlds using text prompts or uploaded images.
The system predicts physics and character interactions, allowing for immediate navigation. It also features a “remixing” capability to modify existing generations. The prototype is currently available to Google AI Ultra subscribers in the US, with sessions limited to 60 seconds.