Google DeepMind just turned two decades of Street View photography into something far more ambitious than helping you find a parking spot. Genie 3, the latest iteration of the company’s generative world model, can now ingest Street View imagery and produce interactive, navigable 3D simulations grounded in actual locations around the globe.

The system draws on a library of 280 billion images captured across 110 countries, effectively giving the model a visual memory of much of the planet’s surface.

What Genie 3 actually does

Feed the model a Street View image of, say, a neighborhood in Tokyo or a stretch of highway in rural Montana, and it generates a simulated 3D environment you can move through in real time. The output runs at 24 frames per second at 720p resolution.

Genie 2 could generate playable environments from text prompts and synthetic images. Genie 3 anchors those generated worlds to places that actually exist.

Waymo is already using Genie 3 to simulate rare driving scenarios for its autonomous vehicle training pipeline. The kind of edge cases that are dangerous to stage in real life, like a cyclist running a red light at an unusual intersection, can be reconstructed from Street View data and varied endlessly by the model.

The path from research project to product

Google DeepMind released a research preview in August 2025, and a consumer-facing version followed in January 2026. The Street View integration announced on May 19, 2026, represents the most significant capability jump since the project’s inception.

Access to the new features is initially limited to select Google AI Ultra subscribers in the United States. Google has indicated plans for a broader global rollout in the coming weeks, though no specific timeline has been published.

Applications extend well beyond autonomous driving. Google is targeting gaming, robotics training, and education as primary use cases.