Google DeepMind announced one of I/O 2026's most interesting integrations: Project Genie, DeepMind's general-purpose world model, now connects to Street View to build worlds rooted in real places.
How it works
Users tap a pin in Google Maps, pick a U.S. location, choose an aesthetic style — presets like "Desert Sands" or "Stone Age" — and describe a character (a favorite animal, a comic book hero, a stop-motion monster). Genie uses Street View imagery as its starting point and generates an interactive environment consistent with what you'd actually see on that street.
The technical leap: spatial continuity
According to TechCrunch and 9to5Google, the real progress is spatial continuity: turn 360 degrees and the model remembers what was behind you and rebuilds the environment coherently. From there, Genie can keep generating new areas on top of that base without losing the thread of the scene.
What it's actually for
DeepMind cites three concrete applications: educational experiences, gaming, and robotics training. Genie 3 is already powering one of Waymo's simulators, used to train self-driving cars on rare events — tornadoes, animals on the road, unusual pedestrian behavior. Adding Street View helps Waymo prepare launches in new cities worldwide without having to physically drive every street.
Availability and limits
Project Genie with Street View rolls out gradually to Google AI Ultra subscribers (18+, globally — but Street View imagery limited to the U.S. for now). DeepMind is candid about the limits: visual quality is still "video game" rather than photorealistic. It's an experiment, not a finished product. But it's the first consumer world model anchored to a real geographic dataset, and a clear signal of where spatial AI modeling is heading.