Google DeepMind grounds Project Genie world model in Street View imagery

Google DeepMind has introduced a new integration between its Project Genie world model and Google Street View imagery. The capability allows users and AI systems to generate interactive, simulated environments anchored to real geographic locations. Rather than creating purely synthetic worlds, Genie now uses Street View's real-world imagery as a foundation, then layers generative styling on top—enabling scenarios such as underwater variants of actual landmarks or historical reimaginations of real places.

The technical underpinning is Maps Imagery Grounding, an existing technology that grounds AI-generated visuals to geographic data. Users can select a location in the United States via the Maps interface, optionally apply a visual style (Desert Sands, Ocean World, B&W Film, Stone Age), describe a character or scenario, and Genie generates a navigable world consistent with those parameters but anchored to the real Street View data at that location.

The company positioned the feature as relevant to agent and robotics research, noting that Genie has already been used to train AI agents in complex virtual settings and citing Waymo's use of the model for hyper-realistic road environment simulation. Grounding simulation in real geographic and visual data could improve the fidelity of training environments for embodied AI systems learning to navigate actual places.

The Street View integration is now available to Google AI Ultra subscribers ($200 per month) globally, though Street View imagery input is currently limited to locations within the United States. Google indicated plans to expand geographic coverage in the future. Project Genie remains positioned as an experimental research prototype within Google Labs, and the company published documentation acknowledging current limitations alongside the announcement.