Google DeepMind links Street View with Genie world model
▼ Summary
– Google DeepMind connected its Project Genie world model to Google Street View’s 280 billion images, enabling users to explore AI-generated simulations of real locations.
– The feature is rolling out to Google AI Ultra subscribers in the US, with a global expansion planned over the coming weeks.
– Waymo uses Genie 3 to train self-driving cars on rare scenarios like tornadoes or elephants on roads.
– The generated environments are experimental, closer to a video game than photorealistic, and lack physics awareness.
– Genie maintains spatial continuity, remembering what was behind you when turning 360 degrees in the simulation.
Google DeepMind has officially merged its Project Genie world model with two decades of Google Street View imagery, enabling users to roam through AI-generated recreations of real-world locations. The announcement, delivered at the Google I/O developer conference on Monday, represents one of the most concrete applications yet of generative world models when combined with an enormous dataset of actual places.
Project Genie, DeepMind’s versatile system for building interactive environments, now taps into a library of over 280 billion images spanning 110 countries and all seven continents. The result is a tool that allows you to step into a simulated version of, say, a snow-covered New York City block or a sunlit London street and move through it in real time.
From research preview to consumer product
The third iteration of the model, Genie 3, debuted as a research preview in August 2025 as part of Google’s broader effort to integrate AI across its ecosystem. By January 2026, DeepMind had opened access to Google AI Ultra subscribers in the United States. The Street View feature is now being introduced to some Ultra users in the US, with a worldwide rollout expected in the coming weeks.
Jack Parker-Holder, a research scientist on DeepMind’s open-endedness team, described the feature as serving two key purposes. For robotics developers, it offers a way to train agents in simulated environments that accurately reflect real locations. For everyday users, it provides a novel way to explore for enjoyment.
Waymo is already a customer
The robotics use case is already operational. Genie 3 powers one of Waymo’s simulators, where the self-driving car company trains on rare events that would be dangerous or impractical to stage in reality, such as tornadoes or unexpected wildlife encounters on roads. Anchoring these simulations in actual Street View geography adds a crucial layer of realism.
This simulation-to-reality pipeline is becoming a critical challenge in physical AI. Companies like Nvidia and Cadence have been competing to bridge the gap between what robots learn in virtual environments and how they perform in the real world. DeepMind’s strategy of layering generative models over real-world imagery offers a distinctive approach.
Impressive, but far from photorealistic
Diego Rivas, a product manager at DeepMind, cautioned that the Street View integration remains experimental. The generated environments resemble a video game more than a photograph, and the model lacks physics awareness. In one demonstration, a character ran straight through a row of cacti without any reaction.
Parker-Holder acknowledged this shortcoming directly, estimating that interactive world generation lags behind video generation by roughly six to 12 months in terms of accuracy. For comparison, Google’s own Veo model already understands basic physics, and its Nano Banana tool can render perfect text in infographics. Genie has not yet reached that level.
The spatial continuity trick
What does work well, according to Jonathan Herbert, director of Google Maps, is spatial continuity. When you turn 360 degrees inside a Genie-generated environment, the AI remembers what was behind you. It maintains a coherent model of the space instead of regenerating it from scratch with every shift in viewpoint.
Herbert described this spatial awareness as the true breakthrough. Google has spent two decades capturing the world through Street View, and the Maps team has long considered how to build richer models on top of that data. Genie, it seems, provides the answer, or at least the beginning of one.
What comes next
The launch fits within a broader pattern at Google, where the company is steadily weaving AI capabilities into products that already boast massive user bases. Street View’s dataset is a competitive advantage that no other AI lab can easily replicate. Connecting it to a generative world model transforms a passive mapping tool into something far more dynamic.
Whether Genie’s simulated streets will eventually match the fidelity of dedicated game engines or professional video production remains uncertain. For now, the feature stands as a compelling proof of concept, one that suggests a future where the boundary between navigating a map and exploring a living, AI-generated world becomes increasingly blurred.
(Source: The Next Web)
