Google DeepMind Integrates Street View With Genie 3 World Model for Real-World Environment Simulation

Google DeepMind announced today at Google I/O that it is connecting Street View to Project Genie, its general-purpose world model that generates interactive environments. The integration gives Genie 3 access to 280 billion Street View images collected over 20 years across 110 countries and seven continents.

The feature allows users to simulate real-world locations and modify environmental conditions including weather, time of day, and scenarios. According to Jack Parker-Holder, research scientist on DeepMind's open-endedness team, a robot deployed in London could use Genie to simulate rare sunny conditions, or users could visualize a New York City block covered in snow.

Technical Capabilities and Limitations

Genie 3 already powers one of Waymo's simulators to train self-driving cars on rare events like tornadoes. The Street View integration extends this capability beyond the car's point of view to other agents including humans and robots.

Parker-Holder told TechCrunch that current output quality is "video game quality rather than photorealistic" and the models lack physics awareness. In demonstrations, simulated figures passed through solid objects like cacti and bushes. He estimates the technology is "six to 12 months behind video in terms of accuracy and quality."

Jonathan Herbert, director of Google Maps, identified spatial continuity as the breakthrough feature. When users turn 360 degrees in a simulation, the AI correctly remembers and reconstructs the environment behind them. Herbert acknowledged that Genie cannot yet create faithful street reconstructions.

Deployment Timeline

Access begins rolling out today to Google AI Ultra subscribers in the United States, with global Ultra subscriber access coming over the next few weeks. Google first released Genie 3 for research preview in August 2025 and opened access to U.S. Ultra subscribers in January 2026.

Diego Rivas, product manager at DeepMind, characterized the feature as experimental with significant accuracy improvements needed. The research team aims to distribute the capability widely to accelerate development.

What This Means

This integration represents Google's first large-scale application of world models to real geographic data, potentially creating a new category of simulation tools beyond gaming and robotics training. The 280 billion image dataset gives Genie 3 unprecedented real-world grounding, though the 6-12 month gap behind video generation quality and missing physics understanding indicate substantial technical challenges remain. For autonomous vehicle companies and robotics developers, the ability to simulate specific locations under varied conditions could accelerate deployment to new geographic markets, particularly when combined with Waymo's existing rare-event training use case.