Google DeepMind unveils Genie 3, a world model that generates explorable 3D environments in real time

Google DeepMind/Genie world models

Google DeepMind unveiled Genie 3, a world model that turns text prompts into navigable 3D environments at 24 fps and 720p, maintaining consistency for several minutes. Unlike Genie 2, which generated static scenes, Genie 3 enables real-time user interaction and supports environment manipulation like weather and object changes.

• Generates interactive, navigable environments from text in real-time: 24 fps, 720p, multi-minute consistency.
• Major upgrade from Genie 2 (static ~1 minute) to dynamic interactivity and realistic physics simulation.
• Supports promptable world events: alter weather, introduce objects, create counterfactual scenarios beyond navigation.
• Step toward practical agent training with longer, more complex task sequences.