🌍Google DeepMind Releases Genie 4 World Model With Two-Minute Memory

Your game prototypes just got a lot cheaper to build

Google DeepMind just dropped Genie 4, and the headline trick is simple: you type a prompt mid-simulation and the world changes. Change the weather, move the camera, add a person or a object. It happens live. The model is in research preview for trusted testers now, with a wider rollout to Google AI Ultra subscribers and Gemini API early-access partners coming in the next few weeks. Why should you care? If you build games, simulators, or anything that needs synthetic environments, this is the first world model that starts to feel like a tool instead of a demo. Pair it with Gemini as the director, which plans steps and checks frames while Genie renders, and you get a rough blueprint for testing robot scenarios in simulation before they touch hardware. That's the part robotics teams should be reading twice. The numbers: Genie 4 holds a consistent scene for up to two minutes, double Genie 3's limit. It adds real-time audio that matches the world, including music, ambient noise, and dialogue. Genie 3 shipped a little over a year ago, so this cadence is fast. Runway's GWM Worlds 2 landed a couple of months back, and the category is consolidating fast: AMD bought World Labs, Amazon is reportedly in talks for Decart, and Rivian acquired Third Dimension. Watch for two things: whether the two-minute consistency holds up outside demos, and how the Gemini API pricing lands when early access opens.