According to DeepMind's blog, Genie 4 is a multimodal model that generates audio together with the world: music, ambient sound and dialogue. Users navigate with the keyboard and can change the simulation while it runs using text prompts, for example changing the weather, moving the camera or adding a person or object. The company says the model simulates physical scenarios such as aerodynamics and collisions, and characters that move and react naturally.
The model builds on the Gemini family: Gemini Omni grounds appearance and behavior in the real world, and Gemini 3.8 Flash acts as an agent that monitors the scene, breaks goals into steps and checks each frame to verify a step was completed. DeepMind also showed a combination with Gemini 4 Argon, in a simulation of Apptronik's Apollo humanoid robot playing checkers. The company says testers in film, journalism, education, gaming and engineering are trying the model, and enterprise partners are evaluating it for scenario generation.
DeepMind says Genie 4 is still an experimental model: visual details and object permanence can drift over time, especially in longer simulations or after multiple edits, and there are limits in legible text, spatial accuracy and complex physical interactions. All video and audio the model generates will carry SynthID watermarks. Timing for broad availability and pricing were not announced.