Google DeepMind on Tuesday released Genie 4, its latest world model for generating interactive simulated scenes, as a research preview for trusted testers^1,3.

The company said the model can produce more photorealistic worlds than Genie 3, accept natural-language edits while a scene is running, generate matching audio, and use Gemini 3.8 Flash to plan and check frames^1,2,3.

TechCrunch reported that users can change weather, camera angle or add objects mid-run, and that scenes can stay consistent for up to two minutes, twice the prior limit¹. The Verge likewise said experiences last about two minutes, up from one minute on Genie 3³.

Ars Technica said Gemini 3.8 Flash acts as a guiding agent that breaks down goals and evaluates each output frame so the simulation can move through environments without dropping the original context as quickly².

TechCrunch said Gemini is more closely integrated, acting as a director that plans steps while Genie generates visuals, and described a research experiment with Gemini Robotics in which a Gemini 4 Argon model broke tasks into steps that Genie 4 simulated in real time¹.

DeepMind presented world models as useful for robot training and rapid creative prototyping rather than as a finished game engine, Ars Technica reported². The Verge said Google pitched pre-visualization for filmmakers, prototyping game ideas and robot training scenarios³.

Genie 4 is DeepMind’s first world model with native audio, according to Ars Technica, including ambient sound and interactive dialogue meant to adapt as the simulation changes². TechCrunch said the audio can include music, ambient noise and interactive dialogue¹.

The outlets also reported remaining limits. TechCrunch said DeepMind acknowledged that details and objects can drift the longer a run lasts, and that accurate audio and legible in-world text are not solved¹.

The Verge said Google warned that details and object permanence can shift over time³. Ars Technica said worlds still remain consistent only for a couple of minutes before they forget how they started².

Access is limited. TechCrunch said the preview is for trusted testers now, with a rollout in coming weeks to Google AI Ultra subscribers in Project Genie and to early-access Gemini API and Google AI Studio partners¹.

Ars Technica said developer access is planned through the Gemini API and AI Studio with pricing not yet available, and that consumer access will phase in on AI Ultra plans that start at $100 a month, with a waitlist for Project Genie².

The Verge’s Jay Peters wrote that he had not been able to try Genie 4 himself because it is launching first as a research preview³. Techmeme carried that Verge report as its lead item on the launch⁴.