Google DeepMind announced Genie 4 on October 27, 2026. The world model creates interactive simulations with sound and lets users steer scenes through natural-language prompts. For now, it is a research preview for trusted testers, not a tool open to the public.

Genie 4 makes generated scenes navigable and responsive, but access is limited and consistency remains an open challenge.

A world that changes on instruction

Users can move through a generated world and request changes as they go, such as different weather, a new camera angle, or another person or object. Google says the model can also generate ambient audio and interactive dialogue alongside the visuals.

Unlike a video that simply plays, Genie 4 is designed to respond while someone explores a scene. Google says the technology could support interactive experiences and simulated environments for robotics and AI agents. Those are proposed uses, not evidence that Genie 4 has already improved real-world robots.

The preview has clear limits

Google says details and objects can drift during longer sessions, especially after complex prompt changes or movement through a scene. Accurate physical interactions, spatial precision, legible text, and reliable audio remain research challenges.

TechCrunch reported that scenes can remain consistent for up to two minutes. That figure does not remove Google’s warning that details may shift over time. Google’s announcement presents demonstrations, not a public benchmark comparison, so its performance claims are company-reported results rather than independent test scores.

Who can use Genie 4?

Google plans to extend access to Google AI Ultra subscribers through Project Genie and to selected Gemini API and Google AI Studio partners. That expansion is planned, not general availability. The current preview gives Google feedback before a broader release, while ordinary consumers and most developers remain without access.

For consumers, the immediate distinction is access: Google is demonstrating an interactive world model, not opening a new product to the public. Wider access remains planned. Whether the scenes can stay coherent and physically accurate as users interact with them remains an open research question.