Simulate real-world places with Project Genie and Street View

Source: Google Date: 2026-05-19 URL: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/project-genie-expands/

Summary

Google DeepMind’s Project Genie — a general-purpose world model — now integrates Street View imagery to generate interactive virtual environments anchored in real U.S. locations. Users select a Maps pin, optionally apply a style (desert, stone age, underwater), and Genie generates a navigable world from the real-world visual context. The feature uses Maps Imagery Grounding, the same API available to developers for AI visual generation. Access is through Google AI Ultra ($200/month).

Implications

• Agent grounding thread. World models that can generate navigable environments from real-world visual context are directly relevant to robotics and autonomous navigation research. Extending Genie with Street View imagery moves it from a purely synthetic world generator toward a model with grounding in real physical geometry — a meaningful step for sim-to-real transfer.
• Model releases / multimodal thread. The Maps Imagery Grounding API being developer-accessible is the more significant signal than the consumer UI. Teams building location-aware agents (delivery, inspection, field service) can use this to generate training environments or test scenarios without physical site visits.
• Enterprise deployment battleground. Gating this at $200/month AI Ultra is a deliberate positioning choice — Google is selling access to world-model infrastructure as a premium tier rather than commoditizing it. That signals confidence in the capability’s differentiation and sets a price anchor for what real-world-grounded simulation is worth.