Google Unveils Genie 4 Real-Time Multimodal World Simulation Model

Google has announced the launch of its world model, Genie 4, and has begun opening a research preview to invited testers. Users can control the viewpoint and movement using a keyboard, and can also input text to change weather conditions, add characters or objects, or modify the scene while it is running. Google states that the model can simultaneously generate visuals, environmental sounds, and context-dependent dialogue, and can simulate physical interactions such as character movements and collisions; however, these capabilities still need to be tested in broader use cases. Genie 4 integrates with the Gemini series of models. Specifically, Gemini 3.8 Flash continuously observes the generated visuals, breaks down goals into steps, and adjusts plans after scene changes. Google is also exploring combining Genie 4 with Gemini 4 Argon and robotics research to simulate multi-step tasks. Invited testers are currently trialing the model in fields such as film, gaming, education, and engineering. Google plans to subsequently open access to Google AI Ultra subscribers within Project Genie, as well as early partners using the Gemini API and Google AI Studio, but has not yet announced a specific date. Google also notes that Genie 4 remains an experimental model: after long runs or multiple prompt modifications, objects and visual details may drift, and complex physical interactions, audio accuracy, and spatial consistency still require improvement. Videos and audio generated by the model will include SynthID digital watermarks.