Google DeepMind announced Genie 4 on Oct. 27, 2026, and said it had begun rolling out the real-time interactive world model as a research preview for trusted testers. The company says Genie 4 generates a simulated world and its audio together, including ambient sounds, music and dialogue that responds as a scene changes.
In the interactions DeepMind describes, users navigate and control the camera with a keyboard. Text prompts can steer the scene as it unfolds, such as changing the weather or adding a person or object. DeepMind says Gemini adapts those requests to the surroundings and monitors the scene between prompts.
Different Gemini models have distinct roles. According to DeepMind, Gemini Omni helps anchor a simulation’s appearance and behavior, while Gemini 3.8 Flash observes frames and breaks goals into steps to guide events. The company also describes an experiment combining Genie 4 with Gemini 4 Argon to simulate a task; it does not present that experiment as part of the wider access plan.
DeepMind says testers are exploring uses including film pre-visualization and explorable game prototypes. Its rollout plan puts Google AI Ultra subscribers in Project Genie and early Gemini API and Google AI Studio partners next, without a specific access date. The company calls Genie 4 an early experimental model: it says scenes can stay coherent over short sessions, but visual details and object permanence can drift over time, particularly during longer simulations or after complex edits.
Comments