Google has opened Project Genie to U.S. subscribers of its AI Ultra tier, letting them sketch, explore, and remix immersive worlds through a web app. Powered by the Genie 3 model, the prototype generates terrain and physics in real time as users move, while Nano Banana Pro and Gemini handle image prompts and character perspective. Access begins today for users 18+.

The underlying world model predicts how environments evolve, a step beyond DeepMind’s earlier agents that mastered fixed games like Chess or Go. By rendering the path ahead dynamically, Genie 3 supports scenarios ranging from robotics simulation to historical reenactments. Early testers reported vivid creations but noted occasional physics glitches, limited realism, and a 60‑second generation cap, plus custom on‑the‑fly events.

Project Genie serves as a research platform for understanding how people will employ generative world‑building tools in media, design, and AI research. Users can download video captures of their sessions or remix community‑shared prompts for fresh ideas. Google frames the rollout as an experiment, promising broader availability once the model’s fidelity and control improve and share them publicly for collaborative storytelling.