On October 27, 2026, Google DeepMind unveiled Genie 4, its latest world model. Unlike conventional video generators that render a static clip from a single text prompt, a world model creates an environment you can interact with and influence in real time. Genie 4 builds on last year’s Genie 3 by adding synchronized sound, sharper visuals, and the ability to edit scenes on the fly while they are running.
According to reports from TechCrunch and The Verge, the release represents Google’s latest push to lead a fast-moving corner of artificial intelligence. While traditional video models generate finished files, world models act more like real-time game engines powered by AI.
Changing the scene in real time
The headline upgrade in Genie 4 is its live steering capability. While a simulation is running, users can type text prompts to change what happens next. You can adjust the camera angle, alter the weather, or insert new people and objects into the scene without starting over from scratch. Diego Rivas, an engineer at Google DeepMind, told TechCrunch that this control is especially helpful during creative ideation and storyboarding, helping creators stay in a state of flow.
Genie 4 also introduces real-time audio generation. Instead of producing silent video clips, it generates background music, ambient environmental noise, and character dialogue synchronized with the visual environment.
Under the hood, Google’s Gemini models act as a director for Genie 4. TechCrunch reported that in experiments, Gemini breaks down tasks into step-by-step instructions, which Genie 4 then turns into visual simulations. The Verge noted that Genie 4 uses Gemini 3.8 Flash to help manage these interactive experiences. The model can now hold a consistent scene for up to two minutes, double the one-minute limit of Genie 3.
From film storyboards to robot training
Google is pitching Genie 4 to several distinct groups. For filmmakers, early testers are exploring it as a rapid prototyping tool to draft storyboards and visualize scenes before shooting. For educators, Google suggested interactive learning experiences where a user might step inside a historical scene or painting and speak with its subjects.
The model is also being used in robotics research. In experiments tied to Gemini Robotics, researchers use Genie 4 to simulate physical tasks in a virtual space before testing them on real hardware.
However, Google acknowledges that the technology still has limits. Both TechCrunch and The Verge noted key flaws highlighted during Google’s press briefing: background details and objects can drift or disappear the longer a simulation runs. In addition, text rendered inside the generated worlds is often unreadable, and audio accuracy remains inconsistent.
How to access Genie 4 and what it costs
Genie 4 is not widely available to the general public today. It is currently in a research preview stage restricted to select trusted testers.
Google plans to roll out Genie 4 in the coming weeks to subscribers of Google AI Ultra through Project Genie, an experimental platform for generating interactive environments. Developers will also gain access through Google AI Studio and the Gemini API. Google has opened a waitlist for users who want to try it in Project Genie.
While full pricing details for Genie 4 access have not been announced beyond requiring an AI Ultra tier, Google’s standard Gemini assistant offers a free plan with basic image editing tools.
Google is far from the only company building interactive environments. TechCrunch noted that startup Runway released its own playable world model, GWM Worlds 2, a few months ago. If you want to compare standard text-to-video tools that are available right now, see our guide on free AI video generators.