Google DeepMind has announced Genie 4, a world model that generates real time interactive simulations with native audio, including ambient sound and interactive dialogue, alongside the visuals. Users can steer a simulation with standard keyboard controls and modify it in natural language as it runs, changing the weather, moving the camera or introducing a new person or object, with the model adapting the scene around the change.
DeepMind said the advance rests on pairing Genie with the Gemini family. Gemini Omni anchors the simulations in real world appearance and behaviour, and a Gemini 3.8 Flash model acts as a guiding agent in the loop, observing the scene, breaking goals into steps and checking that each step succeeded as the world progresses. The company highlighted more natural human movement as the most dramatic realism gain, along with simulation of physics such as aerodynamics and collisions.
The stated purpose is practical as much as creative. DeepMind framed coherent, variable simulations as training environments for robots and agents, and said testers in film, journalism, education, gaming and engineering are using the preview for pre-visualisation, historical reconstruction and rapid prototyping. All video and audio generated by Genie 4 will carry SynthID watermarking.
The limits are stated plainly in the announcement. Visual details and object permanence can drift over longer sessions or after complex edits, and complex physical interactions, real time audio accuracy, legible text and spatial accuracy remain open problems. Genie 4 is rolling out as a research preview to trusted testers first, with Google AI Ultra subscribers in Project Genie and early access partners of the Gemini API and Google AI Studio next in line.