Google DeepMind unveiled Genie 4, a research-preview world model that creates steerable interactive scenes with generated audio in real time.
Google DeepMind has announced Genie 4, a world model designed to generate and simulate interactive scenes in real time. The company says users can guide a scene with natural-language instructions, including moving the camera, changing weather or adding characters. The system also generates audio—including ambient sound, music and dialogue—as a simulation unfolds.
Google describes Genie 4 as combining components from its Gemini model family. Gemini Omni is used to anchor simulations in real-world appearance and behavior, while other Gemini models contribute reasoning intended to help scenes remain coherent and continuous. The announcement focuses on an interactive system rather than a tool that only produces a fixed video: users can intervene and steer the simulated environment as it runs.
The company points to several possible uses. Simulated environments could provide training settings for robots and AI agents, and the technology could support creative work in areas such as filmmaking, journalism, gaming and manufacturing. These are prospective applications; Google’s announcement does not establish that Genie 4 is already a validated training platform or a generally available production tool.
Access is limited for now. Google says it has begun rolling out Genie 4 to a small group of trusted testers. The company plans to bring it to Google AI Ultra subscribers through Project Genie and to early-access partners through the Gemini API and Google AI Studio. The staged release means developers and researchers should treat the announcement as a research preview, not an open launch.
The release adds to industry interest in world models: systems that represent and generate environments users or agents can interact with, rather than responding only with text or static media. If the approach proves reliable, interactive simulations could offer a way to explore scenarios and test behaviors in controlled settings. But the practical value will depend on factors the announcement does not yet resolve, including the fidelity and consistency of generated worlds, the degree of control users can exercise, and how well simulated conditions transfer to real-world tasks.
For now, Genie 4 is an early research effort with limited availability. Google’s announcement highlights its real-time scene steering and native audio, while broader claims about robotics, agents and creative production remain potential applications to be evaluated as access expands.