Google DeepMind Links Street View to Genie to Create Interactive Real-World Simulations

What happened

At the Google I/O 2026 developer conference, Google DeepMind announced the integration of Street View data with Project Genie, the company’s general-purpose world model. This connection enables the platform to transform static imagery from more than 110 countries into interactive, traversable environments that can be manipulated through AI prompts.

What's new in this update

The integration introduces the ability to simulate environmental variations within real-world locations. Users and researchers can now adjust the weather or time of year for a specific street, such as visualizing a Victorian neighborhood in London during rare sunny intervals or a New York City block under heavy snowfall. This moves Genie beyond purely synthetic game worlds into simulations anchored to real-world geography.

Key details

Google’s Street View archive contains more than 280 billion images collected over two decades. By feeding this dataset into Genie 3, the model can simulate worlds from the perspective of different agents, including humans and robots, rather than being limited to a car's point of view. While current outputs are described as having 'video game quality' rather than photorealism, the sheer scale of the data allows for massive environmental diversity.

Background and context

Project Genie was released for research preview in August 2025, and access was expanded to Google AI Ultra subscribers in early 2026. The model is already utilized by Waymo to train self-driving cars on 'exceedingly rare events' such as tornadoes. Integrating the global Street View library is expected to help autonomous systems prepare for expansion into new cities and continents by simulating local conditions before physical deployment.

What to watch next

The feature is rolling out to Ultra users in the U.S. immediately, with global access expected within the next few weeks. DeepMind researchers caution that the model is still an experiment and is not yet 'physics-aware,' meaning it does not yet understand cause and effect, such as objects colliding with or passing through obstacles. Future development will focus on improving physical accuracy and visual fidelity.