When Google first showed off Project Genie, it felt like a cool tech demo: an experimental “world model” that could spin up little playable environments from a text prompt. Now it feels like the start of an entirely new medium. With Street View integration and a broader rollout to Google AI Ultra subscribers, Genie is no longer just inventing worlds from scratch – it is literally rebuilding the real one in AI.

In other words, Google just turned billions of Street View photos into a kind of living, navigable game engine.

If you’ve missed the slow-burn story of Project Genie so far, here’s the short version: it’s Google DeepMind’s “general-purpose world model,” a system that can look at images or text and generate an interactive 2D or 2.5D environment you can actually move around in. Think of those side-scrolling or top-down games that are generated on the fly, but instead of painstakingly crafted levels, Genie is painting the world frame by frame as you explore.

The project has been living inside Google Labs as a research-y playground – technically a prototype – with early access tied to the company’s highest-end AI subscription. Earlier versions could already transform rough sketches, images or prompts into little “playable” scenes, and last year’s Genie 3 update pushed that further with real-time video generation that responds as you move and interact. But even then, it was mostly untethered fantasy, inventing physics and scenery from an abstract idea rather than any particular place.

Street View changes that.

The new update connects nearly 20 years of Google Street View imagery to Genie using what the company calls Maps Imagery Grounding. Under the hood, Genie can now condition its interactive scenes on Street View’s vast dataset – more than 280 billion images from over 110 countries – and use that as the base layer for whatever world you want to explore.

In practice, the experience feels almost disarmingly simple. Inside the Project Genie interface, there’s now a little Maps-style pin. Tap it, pick a place in the United States – say, your childhood street, Times Square, or a random coastal road in California – and Genie loads Street View imagery for that area as the starting point. You choose a visual style like “Desert Sands”, “Stone Age”, “Ocean World” or “black-and-white film,” describe a character (your dog, a claymation monster, a comic-book hero), and hit go.

What appears isn’t just a static panorama. You can walk around, turn 360 degrees, and watch as the world morphs and extends beyond what Street View literally captured. The AI remembers what’s behind you, respects the rough layout of streets and buildings, and keeps the scene coherent as you move – even if the details have been stylized into some surreal, game-like interpretation.

Right now, the Street View integration is limited to U.S. locations, but the plan is to expand to more places over time. Even with that constraint, the canvas is already huge: it’s basically every public Street View spot in the country, turned into a promptable simulation layer you can remix.

On the subscription side, Google is using this moment to push Project Genie out of the “early-access for a few” phase and into its broader AI Ultra offering. Project Genie – including the new Street View capability – is now rolling out to all eligible Google AI Ultra subscribers, a tier that sits at the very top of Google’s AI plans.

Google recently reshaped its AI subscriptions into a three-tier structure: AI Plus, AI Pro, and AI Ultra, with Ultra aimed squarely at heavy users, developers, and teams who want the highest limits and access to experimental features. The flagship Ultra tier (priced around $200 per month for the legacy top plan, with newer plans now starting closer to $100) comes with premium model access, higher usage caps, and add-ons like big chunks of cloud storage and YouTube Premium. Project Genie is one of the perks that helps justify that price: it’s not just another chatbot; it’s an entire interactive medium that only Ultra subscribers can play with, at least for now.

Google says the Genie rollout is global for eligible AI Ultra subscribers aged 18 and over, but the Street View grounding is still U.S.-only to start, which creates an interesting split: global Ultra users can access Genie as a product, but the most futuristic part of the experience is geographically fenced by where Street View imagery is currently enabled inside the prototype.

What makes this release feel bigger than just another feature drop is the way it reframes Street View itself. For almost two decades, Street View has been a kind of interactive slideshow – astonishing in its scope, but ultimately a passive, click-to-advance photo browser. Genie turns that dataset into something more like a simulation substrate, a world that can be reskinned, rewired, and reimagined on demand.

Developers and creators immediately see the possibilities. A robotics team could mock up training environments that mirror real neighborhoods without exposing physical hardware to unpredictable streets. Game designers might prototype levels by starting from real locations – a particular corner in Brooklyn, a mountain pass in Colorado – then layering their own physics and mechanics on top. Educators could drop students into AI-reimagined versions of historical sites, blending real geography with stylized time periods or visual themes.

For everyday users, it’s more about curiosity and self-expression. You can walk through your own city rendered as a prehistoric landscape, or turn your favorite vacation spot into a cel-shaded comic book panel you can actually wander around inside. That blend of familiarity and transformation is surprisingly powerful: the world feels both recognizably yours and completely alien at the same time.

Of course, this is still very clearly a research prototype. Google repeatedly stresses that Genie is experimental, and you can see the rough edges. Visual fidelity isn’t photoreal; the environments often look more like a stylized game than an exact reconstruction, and physics can behave in slightly dreamlike ways. Objects may warp or “breathe” as you move, and details can shift in ways that remind you there is an active generative model behind every frame, not a fixed world.

There’s also a broader conversation here about what it means to simulate real places at this level. Street View has always raised questions about privacy and consent, from blurred faces and license plates to sensitive locations. Grounding Genie in Street View doesn’t automatically make those concerns worse – the system is working off the same imagery Google already hosts – but it does open up new scenarios.

When an AI tool lets you turn any Street View location into an interactive environment, remix it with fantastical or potentially sensitive content, and then share clips or experiences from that, you start to wander into new territory. Google frames Genie squarely as an experiment, not a production game engine or social platform, which gives it room to iterate on safeguards before anything truly mainstream emerges. But if you zoom out a bit, you can see the direction of travel: companies are beginning to treat their geospatial archives not just as maps, but as raw material for synthetic worlds.

It is also worth noting what Genie isn’t trying to be, at least for now. This is not a self-driving car simulator or a precision digital twin of the world; it is more like a vivid daydream anchored loosely to reality. The model is allowed to be wrong about rooftop heights, street textures, or the exact lineup of parked cars, as long as the experience feels consistent and engaging. That trade-off between strict accuracy and playful plausibility is part of what makes Genie fun, but it also keeps it firmly in the “creative tool” bucket rather than anything safety-critical.

The bigger strategic play is how nicely Genie fits into Google’s evolving AI stack. On one side, you have text-and-image models like Gemini for chat, search and productivity. On another, there are coding and agent tools like Google Antigravity, pointed more at developers and automation. Genie adds something different: a way to generate and inhabit continuous, interactive spaces instead of just documents or answers.

By locking that experience to AI Ultra, Google effectively turns its wildest experiments into a perk for its highest-value subscribers. If you’re paying for Ultra, you aren’t just getting a faster or bigger model; you’re getting access to modes of interaction that feel qualitatively new. That’s the kind of differentiation Google needs if it wants its AI bundle to stand out in a market where “chat with a powerful model” is slowly becoming table stakes.

The Street View integration is a particularly strong example of Google leaning into its own unique assets. OpenAI doesn’t own a global mapping fleet. Neither does Anthropic. Google does – and that imagery is now feeding a generative system that can turn the company’s mapping legacy into something more playful and immersive.