Google merges Genie 3 world model with Street View to build interactive simulations of real places

Google DeepMind has expanded its Project Genie world model with data from Google Street View, enabling the creation of interactive, navigable simulations of real places on Earth. Instead of generating random, fictional environments, Genie 3 can now ground its simulations in real photos of streets, squares and buildings collected by Street View cars over more than fifteen years.

How the new feature works

The mechanics are simple: the user drops a pin on any location in the United States on Google Maps, optionally selects a visual style for the world, and then describes the character they want to control, which could be a favorite animal, a comic-book superhero, or a stop-motion-style creature. The system combines Genie's generative power with real Street View imagery, so the generated world preserves the street layout, buildings and landmarks of the actual place.

This marks a significant shift from earlier versions of Genie, which built environments largely from scratch based on a text or image prompt. Anchoring the model in real geographic data means the generated world matches actual topography, which matters both for entertainment and for technical applications.

Data scale and technical applications

The database underpinning the new feature is enormous. Street View has been collecting photos since 2007, and Google says it now has more than 280 billion images from over a hundred countries. Combining that database with the Genie 3 world model creates a tool that goes beyond entertainment: it can serve as a training environment for AI agents and robots that need to learn to navigate complex, realistic surroundings.

Google says Genie 3 has already powered one of the simulators used by Waymo, the autonomous vehicle company. Waymo uses it to train vehicles on rare, dangerous, or hard-to-stage-in-reality scenarios, such as a tornado or an unexpected encounter with an elephant on the road.

Availability and business model

Project Genie, along with the new Street View feature, is rolling out gradually to all eligible Google AI Ultra subscribers worldwide, provided they are 18 or older. A Google AI Ultra subscription costs $200 a month and includes Google's most advanced AI features, including full access to Gemini models and generative tools. Placing Genie in this top-tier, most expensive plan signals that Google treats interactive AI-generated worlds as a premium product rather than a free add-on.

For now, the feature covers locations in the United States, though the Street View photo database is global, suggesting further geographic expansion is possible in future updates.

What this means for the market

Combining world models with real geographic data is a direction other AI labs are also pursuing with so-called world models, systems meant to understand physics and space well enough to train robots and autonomous agents without costly real-world testing. For game and simulation developers, it signals that the line between maps, training simulations, and interactive entertainment is blurring faster than it was just a year ago.

For users in Poland, the feature remains geographically unavailable for now, since it only covers locations in the US, but the underlying architecture, in which a generative model operates on real street imagery, raises the question of when similar tools might extend to European cities mapped by Street View.