DeepMind's Project Genie Enters the World Model Race

Google DeepMind's Project Genie is no longer just a research curiosity. Starting today, Google AI Ultra subscribers in the US can access the experimental prototype, letting users create, explore, and remix interactive worlds using text prompts and images.

But this isn't just a product launch—it's a bid for dominance in one of AI's most contested new frontiers: world models.

According to the DeepMind blog, Project Genie is powered by Genie 3, which the company calls "the first real-time interactive world model capable of generating persistent 3D environments at 24 fps." That's notable because the competition is heating up. Fei-Fei Li's World Labs launched Marble as a competing product, positioning itself as a direct rival in the race to build AI systems that can simulate and generate interactive environments.

The significance: world models simulate how environments evolve and respond to actions—a capability that could matter for everything from robotics to entertainment to training AI systems. DeepMind says Genie 3 can model "any real-world scenario—from robotics and modelling animation and fiction, to exploring locations and historical settings."

The prototype integrates with Nano Banana Pro for image previewing and Gemini for underlying AI capabilities. Users can define their perspective—first-person or third-person—before entering a generated environment.

It's early. DeepMind acknowledges limitations: generated worlds might not look completely true-to-life, and character control has room for improvement. But the company says it plans to expand access over time.