Google DeepMind’s Genie 4 research preview creates steerable simulated worlds with generated audio, while testing remains limited to trusted users.
Google DeepMind announced Genie 4 on October 27, 2026, introducing a new research-preview world model that generates interactive simulations. Users can steer a scene as it unfolds by issuing natural-language prompts—for example, changing the weather, moving the camera, or adding characters. The model also generates audio, including environmental sounds, music, and interactive dialogue.
The system draws on Google’s Gemini models alongside Genie’s world simulation. Google says Gemini Omni contributes to the simulations’ appearance and behavior, while other Gemini models help with reasoning and continuity. In a reported research experiment, a Gemini 4 Argon model broke a task into steps that Genie 4 simulated in real time, illustrating a potential way to test robot scenarios in virtual environments.
TechCrunch reported that Genie 4 can keep a scene consistent for up to two minutes, twice the predecessor’s limit, and described improvements in physics and character realism. The early system has notable shortcomings: details and objects may drift during longer sessions, and Google’s team said accurate audio and legible text inside generated worlds remain challenges. The company has not yet established the final product experience.
Access is currently limited to trusted testers. Google said it plans to bring Genie 4 to Google AI Ultra subscribers through Project Genie and to early-access partners using the Gemini API and Google AI Studio. The work could have applications in areas such as robotics training, gaming, and creative prototyping, but the announcement describes research-stage capabilities rather than a broadly available production tool. Its practical value will depend on whether the team can improve simulation consistency and make its generated environments reliable enough for sustained use.