Genie 4 turns a text prompt or a still image into a world you can explore and change in real time, for teams working on robot and agent training and on interactive media. Keyboard controls handle movement and camera, and natural language prompts change the weather, shift the view, or add people and objects while the simulation continues. Google DeepMind opened the model on October 27, 2026 as a research preview for trusted testers, with access planned next for Google AI Ultra subscribers in Project Genie and for early partners on the Gemini API and Google AI Studio. It is the first DeepMind real-time world model to generate audio with the picture. Ambient sound, music, and dialogue are produced as the scene plays, and characters can speak and react to prompts or to events around them. Gemini Omni anchors how places and people look and behave, while Gemini 3.8 Flash works as a guiding agent that splits goals into steps and checks frames so the simulation stays coherent. DeepMind also describes stronger physics, including aerodynamics and collisions, and people who move and respond more naturally than in earlier Genie models. TechCrunch reports that a scene can hold together for up to two minutes, double the limit of Genie 3. DeepMind has paired Genie 4 with Gemini 4 Argon in robotics research, so agents can rehearse tasks in simulation before those tasks are tried on physical hardware. The lab points to the same kind of controllable world as a training ground for robots and agents, and as a tool for interactive work in filmmaking, games, journalism, and manufacturing.