Google DeepMind unveiled a new extension to its Project Genie platform that anchors the generative world model to real‑world imagery from Google Street View. By linking the model’s procedural terrain creation to actual map data, developers can now place AI agents inside recognizable locations while retaining the sandbox’s creative flexibility and supports seamless export to standard 3D formats for downstream pipelines.

Since its debut, Genie has powered Waymo simulations and other research that demand hyper‑realistic road environments. The new Maps Imagery Grounding layer lets users pick a U.S. pin, choose a visual style—such as “Ocean World” or “B&W film”—and describe a character, prompting Genie to render a scene rooted in the selected Street View backdrop. This capability also enables educators to craft immersive field trips without leaving the classroom.

Access opens today for Google AI Ultra subscribers paying $200 a month, with the Street View feature rolling out globally to eligible users over 18. Though still labeled an experimental prototype in Google Labs, the integration demonstrates how large language‑driven world models can bridge synthetic and real environments, giving robotics and gaming teams a ready‑made testbed anchored in reality. Testers report smoother navigation and richer visuals in early trials.