Google has unveiled Genie 4, a new AI model designed to generate realistic three-dimensional environments. This release expands on the company's series of world models, which began in 2024. The updated system improves upon its predecessor by rendering fluid movements and complex details, while also generating synchronized audio.
Users can build virtual worlds through natural language prompts, allowing for real-time navigation and interactive edits. The model maintains photorealism and consistency when users request changes to weather or object placement. Interface options include both natural language inputs and keyboard controls for camera manipulation.
The architecture integrates several Google models to enhance output. Gemini Omni is utilized for visual realism, while Gemini 3.8 Flash ensures consistency across simulation frames. DeepMind is currently exploring applications in robotics by pairing Genie 4 with the Gemini 4 Argon large language model.
The robotics initiative aims to enable engineers to simulate complex tasks, such as robotic soldering, using high-level instructions. Argon translates these requests into detailed prompts for Genie 4, creating synthetic training data. This process is intended to improve the quality of AI models developed for robotic control.
Google is currently conducting testing with a limited group of early adopters. The company intends to release Genie 4 to the public through its Google AI Ultra subscription, the Gemini API, and Google AI Studio.
