Google has unveiled a new AI experiment that feels more like a research milestone than a product launch. Called Project Genie, the tool comes from Google DeepMind and allows users to create fully interactive virtual worlds generated by artificial intelligence.
The project is now available to Google AI Ultra subscribers in the United States who are over the age of 18. Users can access it through Google Labs, provided they’re subscribed to Google’s top-tier AI plan, which currently costs $249.99 per month.
While the experience looks playful at first glance, Google is positioning Project Genie as something far more serious: a foundational step toward building AI systems that can reason about, interact with, and adapt to complex environments—capabilities widely viewed as essential for artificial general intelligence, or AGI.
From Prompts to Playable Worlds
What sets Project Genie apart from typical generative AI tools is interactivity. Instead of simply generating images or static scenes, the system creates living environments that respond to user actions in real time.
Using text prompts, users can generate a world and then create a character to explore it. Movement is controlled with simple keyboard inputs, and the environment reacts dynamically as the character navigates through it. In demonstrations shared by Google, users build small interactive experiences and even simple games within minutes.
Early testers have already shown how easily the tool can be pushed into familiar territory. Some have recreated basic versions of classic game concepts, building worlds reminiscent of titles like The Legend of Zelda or Super Mario. While these are not full-fledged games, they highlight how quickly Project Genie can turn an idea into a playable simulation.
Google says the worlds are generated at 720p resolution and 24 frames per second, which is notable given that everything—from terrain to object behavior—is produced by AI rather than handcrafted code.
An example of a virtual world created by Project Genie. Credit: Google
Not a Game Engine—At Least Not Yet
Despite the excitement, Google is careful to frame Project Genie as a research experiment, not a consumer product or game development platform. The tool doesn’t offer traditional goals, progression systems, or publishing tools, and Google hasn’t indicated plans to commercialize it in its current form.
Still, the implications are hard to ignore. Building interactive environments normally requires teams of designers, programmers, and artists working for months or years. The fact that an AI system can generate believable, interactive worlds in near real time points to a future where simulation becomes dramatically cheaper and faster.
Users can also remix and modify their creations repeatedly, experimenting with different layouts, objects, and interactions. The environments include basic physics, meaning objects react in realistic ways when touched, moved, or collided with.
The World Model Behind the Scenes
At the heart of Project Genie is Genie 3, an experimental “world model” developed by Google DeepMind. The system also draws on other Google models, including Gemini 3 and Nano Banana, to help generate visuals and structure interactions.
World models represent a growing area of AI research. Unlike models that simply predict text or images, world models aim to understand how an environment evolves over time—and how actions change outcomes within it. This includes cause-and-effect reasoning, spatial awareness, and the ability to anticipate future states.
Google first hinted at Genie 3 in mid-2025, describing it as “a key stepping stone on the path to AGI.” With Project Genie, the company is now letting users interact directly with that research.
According to Google, Genie 3 allows AI agents to predict how a world changes in response to actions, making it possible to explore an unlimited range of realistic environments. That capability, Google argues, is essential for building systems that can reason, solve problems, and eventually take meaningful actions in the real world.
Why This Matters for AGI
Artificial general intelligence is often defined as AI that can perform most tasks as well as—or better than—the average human, with minimal supervision. Achieving that requires more than language skills. An AGI system must understand context, adapt to new situations, and reason about the physical and social world.
Project Genie offers a glimpse into how researchers are approaching that challenge. By training AI to simulate environments and interact with them, Google is experimenting with a form of intelligence that goes beyond answering questions or generating content.
While AGI remains a long-term goal rather than an imminent reality, projects like this show how AI research is shifting away from static outputs and toward dynamic, embodied understanding. Whether through virtual worlds today or real-world systems tomorrow, the ability to model environments may prove to be one of the most important building blocks in the next phase of AI development.
For now, Project Genie stands as an intriguing experiment—one that blurs the line between play, simulation, and serious research into the future of intelligence.
Don’t miss out on our latest news—follow us for the latest AI news, breakthroughs, and insights that matter.