Google DeepMind has introduced Genie 4, a world model currently in research preview. The system generates interactive virtual environments with synchronized real-time audio, including music and environmental sounds. Compared to its predecessor, Genie 3, the new model doubles the duration of consistent scene generation to two minutes and improves physical accuracy.
The model integrates with Google's Gemini, which acts as a director to plan steps and verify frames. In experiments, this combination allows for simulating scenarios before testing robots in the real world. Users can modify scenes in real-time by adjusting weather, camera angles, or adding objects through text prompts.
The technology is targeted at gaming, interactive learning, and robotics. Early testers have identified film prototyping as a potential use case, allowing creators to quickly iterate on storyboards. Google also notes that the model currently faces challenges with detail consistency over long durations and maintaining clear text within the generated environments.
Access to the research preview is currently limited to trusted testers. In the coming weeks, Google plans to expand availability to AI Ultra subscribers, users of the Genie project, and early partners utilizing the Gemini API and Google AI Studio.
