Google DeepMind has opened a research preview of Genie 4, the newest version of its world model, which turns prompts into interactive simulated environments. Access starts with a small group of trusted testers. Google AI Ultra subscribers using Project Genie are due to follow in the coming weeks, along with early-access partners working through the Gemini API and Google AI Studio.
The headline additions are sound and steering. Genie 4 produces music, ambient noise and interactive dialogue by itself, and a user can type new instructions while a scene is running to change the weather, shift the camera or bring in people and objects. Scenes hold together for up to two minutes, twice what Genie 3 managed. Realism leans on Gemini Omni, the video generator introduced in May, while Gemini 3.8 Flash helps keep frames consistent.
Google presents the model as more than a media toy. In a Gemini Robotics experiment, Gemini 4 Argon broke a job into steps and Genie 4 simulated each one live; one example was a robotic arm soldering car parts. The company calls world simulation a crucial step toward AGI and a safe training ground for robots and agents. Other pitched uses are gaming, interactive learning and film prototyping, and everything the model outputs will carry SynthID watermarks.
DeepMind is candid about the limits. Details and objects drift over longer runs, audio accuracy is not solved and neither is legible text inside a world. The team says live steering suits storyboarding and fast ideation, but admits the finished product is not yet clear, and that using Gemini Omni to turn a prototype into final film assets is a distant prospect.
About Genie 4
Genie 4 is Google DeepMind's world model for real-time interactive simulation, announced as research on 27 October 2026. It builds on capabilities from the Gemini family.
It is the fourth model in a line DeepMind began in March 2024. The original Genie made only two-dimensional worlds at one frame per second and was trained on video game footage. Genie 2, introduced in December 2024, added three-dimensional worlds but held them for 10 to 20 seconds at 360p. Genie 3, introduced in August 2025, generated interactive worlds in real time at 720p and 24 frames per second, with about a minute of memory. Project Genie, giving Google AI Ultra subscribers access to Genie 3, launched on 29 January 2026.
The release lands in a crowded field. Genie 3 arrived as a preview a little over a year ago, Runway shipped its second world model, GWM Worlds 2, with audio a couple of months back, and the sector is consolidating: AMD is acquiring World Labs, Amazon is reportedly in talks to buy Decart, and Rivian bought Third Dimension.
Sources · 3