Elon Musk Wasn’t Exaggerating! Google Genie 3 Could Let You “Build GTA6 in Minutes”

AI breakthroughs keep emerging, but the real game-changers are often overlooked. On January 30, Google DeepMind unveiled the long-awaited Project Genie—a fun AI tool, yes, but also a crucial step toward Artificial General Intelligence (AGI): a true “world model” experimental prototype.

Project Genie is now available to U.S. Google AI Ultra subscribers aged 18 and above.

At its core, Project Genie is a real-time interactive environment powered by three components: Nano Banana Pro for image control, the Gemini model for language understanding, and Genie 3 for physical simulation. Genie 3 works like human dreaming: by watching millions of object movements, it learns physics rules like gravity and inertia, then creates an immersive virtual world for users to explore and interact with.

The experience is simple: upload a photo turned into a game-style avatar, input the scene and character prompts, and you can explore deserts, cybercities, or even alien planets. With keyboard controls, the system predicts and renders paths and environments in real time, letting users be both player and director.

As an early-stage experimental model, Project Genie has limitations: each session lasts only 60 seconds to prevent logic collapse and control computational costs. Physical simulations are still imperfect, but the experience is already stunning.

Beyond being a high-end “Minecraft,” Project Genie aims to tackle AGI bottlenecks—data scarcity and embodied intelligence. It can simulate billions of environments, allowing robots to accumulate “muscle memory” and transfer skills to the real world. Future applications could include therapy and education, providing safe and controlled immersive experiences.

Despite rough visuals, short sessions, and lack of multiplayer support, Genie opens the door to physically simulated worlds, bringing the dream of AI world models closer to reality.