[I/O 2026] Project Genie anchors its generated worlds in Street View

Google DeepMind connects its Project Genie world model to Google Street View imagery, allowing AI Ultra subscribers to generate interactive 3D environments.

Google DeepMind connects Project Genie to real Google Street View imagery. Genie is a generalist world model, capable of producing varied interactive environments, already used in research to train agents and simulate road scenes at Waymo. By anchoring it to real locations, the model gains a foothold in reality, useful for evolving agents and robots in settings close to the physical world.

Specifically, in the Project Genie prototype, one chooses a location in the United States from a Maps pin, optionally selects a style such as “Desert Sands,” “Stone Age,” or “Ocean World,” then describes a character. Genie then composes a world whose starting point is based on Street View imagery, thanks to Maps Imagery Grounding technology, the very same used by developers. This allows for exploring the Golden Gate underwater or revisiting a Texan site in a 1920s ambiance.

Street View imagery in Project Genie currently covers locations in the United States, with an announced expansion to follow. Access, expanded internationally, is reserved for Google AI Ultra subscribers at $200, who are of legal age, with the tool remaining an experimental prototype from Google Labs.