Google DeepMind has introduced Genie 4, a major update to its AI that builds virtual worlds. Released as a research preview on Tuesday, this new model does not just make standard AI videos.

It creates photorealistic, 3D spaces that you can actually explore and change while they are running.Real-time changes and live audioGenie 4 lets you make edits on the fly. If you want to change the camera angle, make it rain, or add a new person to the scene, you just type it in mid-simulation.

You do not have to stop and restart the process.It also brings sound to these virtual worlds. The AI generates live audio to match what is happening on the screen, like ambient background noise, music, or even digital characters talking and reacting to your inputs.Simulations can now run for up to two minutes, which is twice as long as last year's Genie 3.

Google also promises better real-world physics and more human-like characters. Powered by GeminiTo make this work, Genie 4 teams up with Google’s Gemini AI.

Gemini acts like a director, planning out the logic and steps, while Genie creates the actual visuals. Tejas Kulkarni, a research director at Google, explained during a briefing that the future of AI will involve these hybrid setups, where Gemini writes the text or code and Genie simulates it live.This setup is already being tested for robotics.

In one experiment, a Gemini 4 Argon model planned out tasks that Genie 4 then simulated, showing how companies could safely train robots in virtual spaces before testing them in the real world.Who will use it? The AI simulation market is getting highly competitive.

Runway recently launched its playable GWM Worlds 2, while big companies like AMD, Amazon, and Rivian are acquiring smaller AI startups to get similar technology.Google sees Genie 4 being used in gaming, interactive learning, and robotics. However, Google DeepMind engineer Diego Rivas noted that it could also be a major tool for filmmakers.

By allowing rapid storyboarding, creators can stay in a creative flow without waiting for long video renders.Google already offers Veo for high-definition video clips. Eventually, developers might use Gemini Omni, Google's multimodal model, to turn these Genie 4 prototypes into final creative assets.The catch and how to get itThe technology is still in its early stages.

Rivas admitted that if a simulation runs too long, the world can start to "drift," causing objects to lose detail or disappear. The team is also still working on perfectly syncing the audio and making text readable inside the generated worlds.Currently, Genie 4 is only available to a small group of trusted testers.

In the coming weeks, it will roll out to Google AI Ultra subscribers through Project Genie (which costs $100 a month), along with early access partners via the Gemini API and Google AI Studio.- Ends