Genie, Genie 2 and Genie 3 (short for Generative Interactive Environments) are artificial intelligence models developed by DeepMind that can generate game-like, action-controllable virtual worlds based on textual descriptions, images, or sketches. Built as autoregressive latent diffusion models, they frame-by-frame interactivity without requiring labeled action data for training.
The first version of Genie was introduced in March 2024. Its successor, Genie 2, released in December of that year, expanded these capabilities to generate diverse and interactive 3D environments.
Genie 3 was released in August 2025, with higher-resolution world generations and multiple minutes of visual consistency. In February 2026, Waymo adopted Genie 3 and create a specialized world model for autonomous driving simulation. The system produces synchronized camera and lidar outputs and creates edge cases that real fleets rarely encounter. Other research teams have explored object-centric and causal variants. World models track individual objects and learn interventions in latent space.
On January 29, 2026, DeepMind released Project Genie to AI Ultra subscribers.