Google DeepMind announced Genie 4, its new world model for real-time interactive simulation. The model generates photorealistic environments with advanced physics, native audio and the ability to alter scenes using text prompts during exploration. Genie 4 Google DeepMind represents a significant advance in world models.
What happened?
Released on October 27, 2026, Genie 4 is the successor to Genie 3 and incorporates capabilities from the Gemini family. Gemini Omni anchors simulations in real-world appearance and behavior, while other Gemini models provide reasoning to maintain coherence. Users can navigate with keyboard controls and modify the world in real time, changing the weather, camera or adding characters and objects.
For the first time, the model generates native audio, including ambient sounds, music and interactive dialogues that adapt to actions. Scene consistency reaches up to two minutes, double the predecessor, with better physics simulation such as aerodynamics and collisions, and more natural human characters.
Why does this matter?
World models like Genie 4 are crucial steps toward AGI, enabling safe and realistic training environments for agents and robots. Applications include game development, filmmaking, journalism, manufacturing and robotics simulations. Integration with Gemini allows multi-step event planning and scene continuity maintenance.
The model is in research preview for trusted testers and will soon arrive for Google AI Ultra subscribers in Project Genie, plus partners via Gemini API and AI Studio.
What changes in practice?
Creators can quickly prototype interactive experiences without complex engines. AI developers gain environments to test agents. For the geek public, this points to the future of AI-generated games and virtual worlds, where the user is a co-creator in real time. The addition of audio and dialogue raises realism to a new level.
Official source: Google DeepMind.
Image credit: Google DeepMind — Source: Google DeepMind
By GeekikiBot
