Google DeepMind launches Genie 4, a multimodal real-time world simulation model

DeepMind announced Genie 4, a new world-simulation model that combines photorealistic visuals, native audio and interactive dialogue, and is being rolled out as a research preview for trusted testers Google DeepMind Blog. The model builds on the Gemini family, with Gemini Omni anchoring appearance and behavior while other Gemini models provide reasoning to keep simulations coherent.

Genie 4 lets users steer scenes in real time with natural-language prompts, altering weather, camera angles or adding objects while the system adapts the environment. It simulates physics such as aerodynamics and collisions, and generates ambient sounds and human speech that respond on the fly. Gemini 3.8 Flash acts as a guiding agent, breaking down multi-step tasks and adjusting plans as the world evolves.

Trusted testers across creative, journalism, gaming and engineering fields are using Genie 4 for rapid “draft mode” pre-visualisation, turning archival photos into immersive simulations, and generating zero-to-one scenario prototypes without writing engine code.

Our read: Genie 4 demonstrates a tangible step toward hybrid world models that blend sensory generation with structured reasoning, but its short-session limits and occasional drift mean it remains a research tool rather than a production-ready platform.