Google DeepMind’s Genie 4, announced on October 27, 2026, introduces massive advancements in AI-generated world simulation. Shifting from passive video creation to fully immersive spatial computing, it allows users to generate, navigate, and actively rewrite 3D environments on the fly.
Watch Genie 4 in action to see how real-time prompts transform a static image into a dynamic, explorable world with native sound:
Core New Features of Genie 4
- Native Multimodal Sound and Dialogue — For the first time, Genie 4 generates synchronized audio such as music, ambient environment noises, and interactive dialogue natively alongside the visual simulation, eliminating the need for separate audio generation models.
- Real-Time “Mid-Simulation” Prompting — Users are no longer passive observers. While actively navigating an environment using standard keyboard and camera controls, you can issue natural language text prompts to instantly reshape the world — changing the weather, shifting camera angles, or dropping new characters and objects into the scene seamlessly.
- Gemini-Powered Reasoning & Coherence — Genie 4 builds directly on the Gemini architecture. It integrates Gemini Omni to anchor simulations in photorealistic real-world appearance, while secondary Gemini reasoning models continuously track the environment to maintain narrative, logical, and physical continuity across multi-step interactions.
- Persistent Visual World Memory — The simulation maintains complete environmental consistency for minutes at a time. If you edit a specific part of the world (like painting a wall or adding an object) and look away, the model remembers the state and keeps it intact when you return to that location.
- Gemini API and Developer Availability — Marking a shift from locked research previews, Google is making Genie available in the Gemini API / AI Studio for early-access partners and developers for the first time, alongside an upcoming rollout for Google AI Ultra subscribers.
