Google DeepMind has announced Genie 4, the newest version of its world model, which builds interactive simulated environments in real time and generates sound along with the picture. It is a research preview, and for now only a limited group of trusted testers can use it.
A world model is an AI system that creates a playable, changing scene rather than a fixed video clip. Genie 4 is meant to feel closer to a live simulation, where the person watching can keep changing what happens while it runs.
Sound that reacts to the scene
Google calls Genie 4 the first real-time world model to produce audio natively. Ambient noise, music and spoken dialogue are generated together with the visuals, and the dialogue adapts to user input and events in the world.
The company’s DeepMind write-up also says the model simulates physics such as aerodynamics and collisions, with characters that move and respond to their surroundings.
Changing a world while it runs
Users can type plain-language prompts mid-simulation, for example to change the weather, move the camera or add a person or object. Gemini models adapt each prompt to the scene and try to keep the world coherent between edits. Google says Gemini Omni anchors the output to real-world appearance and behaviour.
Google admits the model has limits. Visual details and object permanence can drift, especially in longer sessions, after several prompt edits or during navigation. Complex physical interactions, readable text and long-term consistency are listed as areas still under research. All generated video and audio carry SynthID watermarking.
Who gets access, and when
Google says it has started rolling Genie 4 out to trusted testers. It is also coming soon to Google AI Ultra subscribers through Project Genie, and, for partners with early access, to Google AI Studio and the Gemini API. A waitlist form is linked from the announcement. Google did not state clip length, resolution or frame rate.
The company frames world simulation as a step towards AGI, pointing to training environments for robots and agents, and to uses in filmmaking, journalism, gaming and manufacturing. No date has been given for the wider rollout.
