Google announces Genie 4 world model for interactive AI experiences

Google DeepMind announced Genie 4, a new world model for real-time interactive simulation. It is designed to produce more photorealistic environments with fluid movement, native audio, and natural-language steering that can change weather, objects, characters, or camera movement while a scene is running. Google says Gemini models help anchor simulations in realistic appearance and behavior and maintain coherence, but the announcement does not identify Gemini 3.8 Flash as the specific model used. Genie 4 is being tested as a research preview by trusted testers, with access planned for Google AI Ultra subscribers through Project Genie and for early-access partners through the Gemini API and Google AI Studio.

Why it matters

Genie 4 could let filmmakers, game developers, educators, and robotics researchers explore interactive scenarios before building them in conventional tools or engines. Its current limitations include shifting visual details and object permanence during longer sessions or after complex prompt edits, along with unresolved issues involving physical interactions, audio accuracy, text rendering, spatial accuracy, and long-term consistency.

Key points

  • Genie 4 can generate explorable, photorealistic interactive environments with fluid movement and native audio.
  • Users can steer simulations in real time with natural-language prompts, including changes to weather, objects, people, and camera movement.
  • Google says the model is being explored for film pre-visualization, interactive experience prototyping, robotics, and other simulation tasks.
  • The official announcement does not substantiate the draft’s claims that Genie 4 uses Gemini 3.8 Flash or supports sessions lasting about two minutes versus one minute for Genie 3.
  • It is currently a research preview for trusted testers, with planned access through Project Genie for Google AI Ultra subscribers and through early-access developer channels; a waitlist is available.

Based on reporting by The original announcement, The Verge, Ars Technica and TechCrunch.