Genie 4 Is Google DeepMind's New World Model With Live Text Steering and Native Audio

Google DeepMind introduced Genie 4, a new world model in research preview that you can steer with text prompts while a simulation runs and that generates its own audio, from ambient sound to dialogue. Google's trusted testers and a DeepMind engineer connect that live control to film prototyping, storyboarding and early ideation.

• Text prompts reshape the scene mid-simulation. You can move the camera, change the weather or introduce a new person or object, and Gemini adapts the prompt to fit the surrounding scene.
• Audio is generated with the world. Genie 4 produces music, ambient noise and interactive dialogue alongside the visuals.
• Scenes stay consistent for up to two minutes. That doubles Genie 3's limit, according to TechCrunch, which also reports better physics and more humanlike characters.
• Gemini models carry the reasoning. Gemini Omni anchors simulations in real-world appearance and behavior, while other Gemini models help keep each simulation coherent and continuous.
• Access starts with trusted testers. Google AI Ultra subscribers in Project Genie and early access partners in the Gemini API and Google AI Studio are next.

Keyboard controls move the camera while Gemini keeps the world on track

In the recorded demos from DeepMind's Genie 4 announcement, users navigate and control the camera with standard keyboard controls and type text prompts to steer the scene as it evolves. Between those inputs, Gemini monitors and steers the scene, according to the post.

According to TechCrunch's report, Gemini acts as the director that plans steps and checks frames while Genie generates the visuals.

"What we’ll increasingly see in the future are these hybrid world models, where you have a combination of Gemini writing code and text, and Genie ingesting that in real time and simulating reality at different levels of abstraction," Tejas Kulkarni, research director at Google, told reporters at a briefing covered by TechCrunch.

The Genie 4 model page says the model generates environments where objects collide, water flows and trees sway in the wind, with generated characters that move and behave naturally. One of its example prompts follows an extravagantly dressed man walking backstage at a musical theatre with a fluid handheld camera.

Example prompts in the DeepMind announcement also read like shot descriptions. One asks the model to use a reference image as the exact first frame, render a low-contrast 16mm documentary look, and let the camera glide freely over a lake around a saxophone player.

Genie 4 joins other real-time world models we have tracked, including Skywork's Matrix-Game 3.0, a 720p model for interactive scene generation.

On the platform side, we also covered fal Worlds, fal's platform for directing AI-generated worlds in real time.

Trusted testers point to film prototyping, and DeepMind cites storyboarding and ideation

The Google DeepMind team frames the model around training environments for robots and agents and new interactive experiences. Google's announcement lists filmmaking among the creative industries it could serve.

Trusted testers are pointing to prototyping films, according to TechCrunch. Google DeepMind engineer Diego Rivas tied the real-time element to early creative work.

"The real time aspect becomes most valuable in the areas where keeping a state of flow in the creativity space becomes incredibly useful," Rivas said during the briefing. "For example, in the ideation phase, in the storyboarding phase."

Rivas added that users could eventually rely on Gemini Omni to turn those prototypes into creative assets, though he said those capabilities are still a ways off.

Scene drift and audio accuracy remain open during the trusted-tester preview

The Genie team acknowledged that details and objects can drift the longer a simulation runs, and that it has not worked out how to keep audio accurate or in-world text legible, TechCrunch reports. Rivas stressed that it is early and that the end-to-end product experience is not yet defined.

Genie 4 is rolling out as a research preview to trusted testers while Google gathers feedback, with consumer access starting with Google AI Ultra subscribers.

Those subscribers will use Project Genie, which also hosts the Genie 4 waitlist.

Developer access begins with early access partners of the Gemini API and Google AI Studio.