Google DeepMind 发布 Genie 4 实时世界模型

Google DeepMind has announced Genie 4, a world model that generates interactive, three-dimensional environments with realistic visuals, sound and dialogue in real time.

The model is currently available as a research preview to trusted testers across creative industries, journalism, education, gaming and engineering.

Genie 4 combines capabilities from Google’s Gemini models to simulate how environments, objects and people behave in response to user inputs. Users can modify scenes through natural-language prompts, change weather conditions, move cameras and introduce new objects or characters while the simulation is running.

“Genie 4 is our first real-time world model to generate audio natively,” the company said in its announcement. The model generates ambient sounds and interactive dialogue, allowing characters to respond to user inputs and events within a simulated environment.

The company said Genie 4 could support applications ranging from robotics and AI agent training to game development, filmmaking and other interactive experiences.

Genie 4 builds on capabilities from Google’s Gemini family. Gemini Omni provides a foundation for real-world appearance and behaviour, while other Gemini models contribute reasoning to maintain coherence across simulations.

Google DeepMind said Gemini 3.8 Flash acts as a guiding agent during complex simulations. It observes the generated environment, breaks tasks into steps, tracks progress and checks whether individual steps have been completed. The model can adjust its plan when the environment changes or users intervene.

The combination could help researchers test how AI systems navigate environments and perform multi-step tasks before deploying them in the real world.

Google DeepMind is also experimenting with Gemini 4 Argon and Genie 4 alongside its robotics research. In one example, Argon breaks down a task into steps that Genie 4 renders in real time, including a simulation of Apptronik's Apollo humanoid robot playing checkers.

The company said filmmakers could use Genie 4 to test scenes by changing camera angles, lighting, dialogue and actions in real time. Game developers could create interactive prototypes from text prompts before writing code for a game engine.

Photojournalists and researchers are exploring ways to turn archival photographs into interactive simulations of historical environments. Other potential applications include visualising events such as flash floods and wildfires.

The company said all video and audio generated by Genie 4 would include SynthID, its digital watermarking technology for identifying AI-generated content across visual and audio formats.