Google DeepMind Unveils Genie 4 With Real-Time Audio and Editable AI Worlds

TL;DR

• Google DeepMind has released Genie 4 in research preview, introducing more photorealistic interactive worlds and real-time audio.
• Users can change weather, camera angles, characters and objects through prompts while a simulation runs.
• The model can maintain a consistent scene for up to two minutes, twice its predecessor’s limit.
• Closer integration with Gemini could support applications in robotics, gaming, education and film prototyping.

Genie 4 Adds Audio and Real-Time Scene Changes

Google DeepMind released Genie 4 in research preview on Tuesday, expanding its world-model technology with synchronized audio, improved realism and more flexible interaction.

The model generates environments that users can explore and modify as they run. Through text prompts, users can change the weather, adjust the camera angle or introduce new people and objects without starting a separate simulation.

Genie 4 also produces audio that responds to the environment, including music, ambient sounds and interactive dialogue.

Compared with Genie 3, which entered research preview a little over a year ago, the new model offers improved physics and more humanlike characters. Google says it can maintain a consistent scene for up to two minutes, doubling the previous model’s limit.

Gemini Integration Opens Possibilities for Robotics

Google has more closely connected Genie 4 with Gemini, allowing the two systems to take complementary roles.

Gemini acts as a director, planning actions and checking frames, while Genie generates the simulated environment.

In a research experiment associated with Gemini Robotics, a Gemini 4 Argon model divided tasks into steps that Genie 4 simulated in real time. The demonstration points to a potential way to test robot scenarios virtually before attempting them in the physical world.

“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,” Google research director Tejas Kulkarni said during a briefing.

Early Testers Explore Film Prototyping

Google identifies gaming, interactive education and robotics as potential applications for world models. Early testers are also exploring their usefulness in filmmaking.

The ability to alter a scene immediately could help creators experiment with ideas, camera positions and storyboards more quickly.

“The real time aspect becomes most valuable in the areas where keeping a state of flow in the creativity space becomes incredibly useful,” Google DeepMind engineer Diego Rivas said.

“For example, in the ideation phase, in the storyboarding phase — anytime you’re kind of in need of rapid ideation.”

Google already offers Veo, a video-generation model that creates high-definition clips with synchronized audio. Rivas suggested that Gemini Omni could eventually help turn interactive prototypes into creative assets, although he emphasized that those capabilities remain some way off.

World-Model Competition Intensifies

Genie 4 arrives as competitors expand their own interactive simulation systems.

Runway introduced GWM Worlds 2 a couple of months earlier, offering a playable world model that responds to user input and generates audio.

The sector is also seeing consolidation. AMD is acquiring Fei-Fei Li’s World Labs, Amazon is reportedly discussing an acquisition of Decart, and Rivian recently bought Third Dimension to support simulation work for autonomous driving.

These developments reflect interest in using world models beyond entertainment, particularly for testing systems and exploring scenarios before real-world deployment.

Research Preview Retains Technical Limitations

Google stressed that Genie 4 remains an early research system, with its eventual product experience still taking shape.

Objects and visual details can drift during longer simulations. The team also acknowledged unresolved challenges involving audio accuracy and the readability of text generated inside environments.

Genie 4 is initially available to trusted testers. In the coming weeks, Google plans to expand access to Google AI Ultra subscribers through Project Genie, alongside early access partners using the Gemini API and Google AI Studio.