Google is offering AI Ultra subscribers in the U.S. a sandbox full of pixels and possibilities with Project Genie—a web-based prototype that lets users create and navigate interactive, responsive worlds in real time. Built on Genie 3, Google’s most advanced “world model” to date, the system dynamically generates environments based on user input, predicting physical interactions and evolving as you explore. This isn’t just a prettier version of The Sims; it’s an ongoing experiment in environment modeling that could eventually inform everything from animation workflows to historical reconstructions.
The tool combines three main modes—World Sketching (where you build your landscape), World Exploration (where you move through it), and World Remixing (where you borrow others’ prompts and add your spin). Users can upload images or simply describe scenes in text, add a character, define a perspective, and even dictate how they travel—on foot, by plane, or otherwise. The system is powered by Google’s Nano Banana Pro and Gemini for finer-grain control over visuals and physics. Yes, they named a model “Nano Banana Pro,” and no, they didn’t explain why.
Project Genie is still very much in beta, with quirks like characters that occasionally don’t respond perfectly and environments capped at 60 seconds of generation. Still, the potential is clear: giving users hands-on access to world models could open up new applications across creative, educational, and even commercial domains. Access is rolling out now to U.S.-based AI Ultra subscribers (18+), with plans to expand. As Google puts it, this is research in action—and their lab is your playground.
Read more at Google Deepmind.