Google DeepMind Introduces Genie 4 World Model

Google DeepMind has introduced Genie 4, a research-preview model that generates interactive virtual worlds in real time and lets users steer them with text prompts. The company says the system adds native audio and more realistic movement while combining with Gemini models to guide events as simulations unfold. Access is currently limited to select trusted testers. Unlike a finished video, Genie 4 generates a scene frame by frame and responds as users move through it or change its conditions. Users can make adjustments such as changing the weather, moving the camera or adding people and objects, according to Google DeepMind. The model is designed to maintain the scene’s context as those changes occur. Google DeepMind says Gemini 3.8 Flash acts as a guiding agent: it observes frames, breaks a goal into steps and checks whether each step has succeeded. It can adapt its plan when a user intervenes or the simulated world changes. The company presents this combination as a way to model complex events and potentially create training environments for robots and other AI systems. The model also generates sound within simulations, including environmental audio and character dialogue,...