Google DeepMind Unveils Genie 4 World Model

Google DeepMind announced Genie 4, a new world model designed to provide unprecedented realism and improve real-time interactive simulation. The system is natively multimodal, capable of simultaneously generating audio-like music, ambient noises, and interactive dialogue within the simulated world. It marks the first real-time world model to generate audio natively, including ambient sounds like wind and waves. The model incorporates capabilities from the Gemini family, using Gemini Omni for real-world appearance and other models for complex reasoning. Specifically, Genie 4 leverages Gemini 3.8 Flash as a guiding agent to break down goals and evaluate each output frame to keep the simulation on track. In this architecture, Gemini acts as the director that plans steps and checks frames while Genie creates the visuals.

The model simulates how the world responds, sounds, and how people move, react, and speak after an event is set in motion. It supports real-world physics scenarios including aerodynamics and collisions. Genie 4 creates characters that move naturally, respond organically to their environments, and can be interactively steered in real-time. Users can use natural language to modify a simulation in real-time, such as changing weather or introducing new objects. The system supports changing parameters midstream, such as weather or camera position, with Gemini helping to steer the simulation. This allows users to type prompts mid-simulation to change the scene, generating what the company describes as the most photorealistic and interactive worlds yet.

Google has started rolling out Genie 4 as a research preview for trusted testers, including Google AI Ultra subscribers and early access partners. The release occurred on Tuesday. However, the technology has limitations regarding consistency. Genie 4 worlds remain consistent for only a couple of minutes before they forget how things started. The model can maintain a consistent scene for up to two minutes, which is double its predecessor's limit.

Tejas Kulkarni, research director at Google, stated that future world models will increasingly be hybrids where Gemini writes code and text while Genie simulates reality. Diego Rivas, a Google DeepMind engineer, noted that the real-time aspect is valuable for ideation and storyboarding phases in film prototyping. The release positions the tool for applications requiring immediate feedback and dynamic environmental changes.