Google DeepMind launches Genie 4 for real-time world simulation

A new interactive world model

Google DeepMind has introduced Genie 4, a world model designed to create interactive, three-dimensional environments. Unlike static video generators, this system produces realistic visuals, sound, and dialogue that respond to user inputs in real time. The model is currently being shared as a research preview with trusted testers in fields like journalism, gaming, and engineering.

The technology allows users to manipulate a scene dynamically using natural-language prompts. For example, a user can change the weather conditions, move the camera, or introduce new characters and objects while the simulation is actively running. This level of interactivity means the environment behaves like a living space rather than a pre-rendered animation.

Google DeepMind plans to expand the availability of this technology to Google AI Ultra subscribers through Project Genie. It will also be made available to early-access partners who use the Gemini API and Google AI Studio. This rollout suggests a move toward making high-fidelity world simulation accessible to a broader range of professional developers.

How the model generates reality

Genie 4 functions by combining the capabilities of various models within the Gemini family. Gemini Omni serves as the foundation for establishing how objects and people appear and behave in the real world. Other Gemini models provide the underlying reasoning necessary to keep the simulation coherent as it evolves.

A key technical feature is the use of Gemini 3.8 Flash as a guiding agent for complex simulations. This specific model observes the generated environment and breaks down large tasks into individual steps. It tracks the progress of these steps and checks if they are completed, adjusting the plan if the environment changes or a user intervenes.

The system also features native audio generation, which is a first for the Genie series. This allows the model to produce ambient sounds and interactive dialogue simultaneously. Characters within the simulation can respond to specific events or user inputs, creating a more immersive and reactive experience for the person interacting with the model.

Applications for robotics and film

One of the primary uses for Genie 4 is training AI agents and testing robotics in simulated environments. Researchers can use the model to see how systems navigate complex spaces and perform multi-step tasks before they are deployed in the physical world. For instance, DeepMind is already experimenting with the model alongside its robotics research to simulate humanoid robot actions.

Filmmakers can use the tool to test out different scenes by adjusting lighting, camera angles, and dialogue on the fly. This allows for a faster iteration process where creators can visualize different versions of a scene without rebuilding the entire set. It provides a sandbox for exploring narrative possibilities before moving into production.

Game developers can also use Genie 4 to build interactive prototypes from simple text prompts. This allows them to test gameplay mechanics and world interactions before they begin the labor-intensive process of writing code for a traditional game engine. It bridges the gap between a creative idea and a functional technical prototype.

Historical and environmental tools

Beyond entertainment, the model has potential uses in education and journalism. Photojournalists and researchers are looking at ways to transform old archival photographs into interactive simulations of historical environments. This could allow people to virtually walk through past events or locations that no longer exist in their original form.

The technology could also be used to visualize environmental hazards such as wildfires or flash floods. By creating a simulated world that reacts to specific variables, researchers can better understand how these disasters impact different terrains. This provides a safe way to model dangerous events for planning and emergency response purposes.

To ensure responsible use of this technology, Google DeepMind stated that all video and audio created by Genie 4 will include SynthID. This is a digital watermarking technology designed to identify AI-generated content across different formats. This measure helps maintain transparency as the model becomes more capable of creating realistic human dialogue and visuals.