The TL;DR
Google DeepMind connected its Project Genie world model to 280 billion Google Street View images, allowing users to explore AI-generated simulations of real-world locations. Announced at Google I/O, the feature is rolling out to Google AI Ultra subscribers in the US, and Waymo is already using Genie 3 to train self-driving cars in rare situations.
Google DeepMind has connected its Project Genie world model to 20 years of Street View imagery, allowing users to navigate through AI-generated simulations of real-world environments. The integration, announced at the Google I/O developer conference on Monday, marks one of the most concrete demonstrations yet of what generative earth models can do when paired with massive real-world datasets.
Project Genie, the company’s general-purpose system for creating interactive environments, can now draw on more than 280 billion images captured in 110 countries and all seven continents. The result is a tool that lets you step into a simulated environment, such as a snow-covered New York City block, or a strangely sunlit London street, and navigate it in real time.
From research preview to consumer product
Genie 3, the latest iteration of the model, debuted as a research preview in August 2025, part of Google’s broader plan to embed AI into its platform stack. In January 2026, DeepMind opened access to Google AI Ultra subscribers in the United States. Street View integration is now open to select Ultra users in the US, with global expansion planned in the coming weeks.
💜 for EU tech
The latest talk from the EU tech scene, a story from our genius founder Boris, and some incredible AI art. It’s free, every week, in your inbox. Register now!
Jack Parker-Holder, a research scientist at DeepMind’s open source team, pitched the feature as serving two different audiences. On the other hand, robotics developers can use it to train agents in simulated environments that mimic real environments. On the other hand, regular users can just check for fun.
Waymo is already a customer
The robot angle is not theoretical. The Genie 3 already powers one of Waymo’s simulators, which the self-driving car company uses to train for rare events that may or may not happen on stage in real life, things like hurricanes or unexpected encounters with elephants on the road. The ability to anchor these simulations in a real Street View environment adds another layer of realism.
This kind of simulation-to-reality pipeline becomes a significant bottleneck in real-world AI. Companies including Nvidia and Cadence have been racing to close the gap between what robots learn inside computers and how they act when deployed, and DeepMind’s method of layering generative models over real-world imagery offers a different route.
Impressive, but far from photorealistic
Diego Rivas, product manager at DeepMind, cautioned that Street View integration remains experimental. The generated environments look closer to a video game than a photo, and the model is still physics-aware. In one scene, the actor ran straight into a row of cacti without any consequences.
Parker-Holder acknowledged the gap directly, estimating that the world’s interactive generation lags behind video production by about six to 12 months in accuracy. In context, Google’s Veo model already understands basic physics, and its Nano Banana tool can provide complete text in infographics. Genie doesn’t exist yet.
A strategy for local continuity
What works best, according to Jonathan Herbert, the director of Google Maps, is the continuity of the area. Turn 360 degrees within the Genie-generated environment, and the AI remembers what was behind you. It maintains a consistent model of the space instead of updating it from scratch with every view shift.
Herbert described this local awareness as a real success. Google has spent two decades capturing the world with Street View, and the Maps team has long been thinking about how to build richer models on top of that data. Genie, it seems, is the answer, or at least one start.
Next
The launch fits a broader pattern at Google, where the company is gradually adding AI capabilities to products that already have large user bases. Street View’s data set is a competitive edge that no other AI lab can easily replicate, and connecting it to a productive world model turns a passive mapping tool into something very powerful.
Whether Genie’s simulated roads will ultimately rival the reliability of dedicated game engines or professional video production remains to be seen. For now, the feature is a compelling proof of concept, pointing to a future where the line between navigating a map and exploring a living, AI-generated world becomes increasingly difficult to draw.
