Google's Genie world model simulates real streets with Street View data

At Google I/O 2026, Google DeepMind unveiled a major integration: Street View imagery is now feeding into Project Genie, the company's general-purpose world model. This allows users to drop into any real-world street captured by Google's fleet and simulate it interactively. You can change the weather, time of day, or even create hypothetical disaster scenarios. The goal is twofold: enable humans to explore places immersively and train AI agents—like robots and self-driving cars—on realistic, diverse conditions. For instance, a robot deployed in London can practice handling the rare sun glint off Victorian housing before it ever encounters it in the real world. Street View's massive dataset—over 280 billion images across 110 countries—provides an unprecedented foundation for world simulation.

However, the integration is still experimental. The simulated worlds are video-game quality, not photorealistic, and the model lacks physics understanding—characters can walk through cacti without collision. Parker-Holder estimates physics will catch up in 6–12 months. The rollout begins today for Ultra subscribers in the US, with global access coming in weeks. Genie 3, which was released last August, is already used by Waymo to simulate rare events like tornadoes or elephant encounters for self-driving car training. Combining Street View with Genie could help Waymo expand to more cities by simulating street-level conditions from a human or robot perspective, not just a car's point of view.