Google unveils Genie 4, a new AI model offering more realistic experiences
Google DeepMind has announced Genie 4, a new world model capable of real-time audio generation and steering via natural language prompts. The model, which can create consistent simulations for up to two minutes, is currently being shared with a limited test group.
On October 27, 2026, Google DeepMind announced Genie 4, a next-generation world model designed to create interactive digital experiences. Offering higher photorealism compared to the previous version, the model allows users to instantly change elements within the simulation, such as weather, camera angles, and objects, using natural language commands.
Genie 4 is capable of generating music, ambient sounds, and interactive dialogues through its real-time audio generation capability. The system, which doubles the simulation duration to two minutes compared to the previous model, is powered in the background by the Gemini 3.8 Flash model. In research experiments, Gemini 4 Argon was used to decompose tasks for robotic scenario testing.
Google is positioning this technology in the fields of film pre-visualization, game prototyping, and robotics training. Currently available only to trusted test users, the model is planned to be released to Google AI Ultra subscribers and early access partners of the Gemini API and Google AI Studio in the coming weeks.
Assessment: Google acknowledges that despite increasing the model's technical capacity, challenges such as object persistence drifts in long-duration simulations, detail degradation, and the readability of in-world text persist.