Google releases Genie 4 preview with sound and real-time world editing

Google DeepMind has released a research preview of Genie 4, a world model that lets users modify generated environments as simulations unfold. Google said the update adds audio matched to the scenes, including dialogue, music and ambient sound.

Access is initially limited to trusted testers. Google plans to expand availability in the coming weeks through Project Genie for Google AI Ultra subscribers, as well as to early access partners of the Gemini API and Google AI Studio.

Google said users can change weather, reposition the camera and add objects or people by entering prompts while a simulation runs. It also described improvements in visual realism, physics and character behavior compared with Genie 3.

The update deepens the connection with Gemini, which plans steps and checks frames while Genie generates visuals. A research experiment associated with Gemini Robotics used Gemini 4 Argon to break tasks into steps for live simulation by Genie 4, suggesting a potential method for testing robot scenarios.

Google is presenting gaming, robotics and interactive learning as potential commercial applications. Trusted testers have also identified film prototyping as a possible use. Google DeepMind engineer Diego Rivas said quick iteration could support creative work such as storyboarding, while emphasizing that the eventual product experience remains uncertain.

The team acknowledged continuing limitations. Objects and details can drift during longer simulations, and ensuring accurate audio and legible text inside the generated environments remains unresolved.