Google DeepMind has made Genie 4, its most advanced world model for generating interactive 3D worlds, available to eligible Google AI Ultra subscribers through Project Genie. The model, which had been a research preview, lets users reshape a world with text prompts while they are exploring it, with no need to regenerate the scene.
Those mid-session edits cover changes such as switching the weather, adding a character or redirecting the action, and Genie 4 rebuilds the visuals and sound around the user in real time. Google AI Pro members can join a waitlist for access.
Genie 4 is also the first version of the model with native sound. Audio adapts to what is happening in the scene, from ambient soundscapes to character dialogue, and a new Project Genie agent can generate a custom background score for a world.
Key facts
- 01Google DeepMind made Genie 4 available to eligible Google AI Ultra subscribers through Project Genie on Nov. 18, 2026.
- 02Genie 4 is the first Genie model with native sound, including ambient audio and character dialogue.
- 03Users can change weather, add characters or redirect action by typing prompts mid-session, without regenerating the world.
- 04Genie 3 sessions now run up to 3 minutes per generation, up from 60 seconds, and worlds can be chained together.
- 05Google AI Pro members can join a waitlist; all output carries SynthID watermarks.
That agent can also build a world from a spoken or conversational description. Gemini can join a session as well, watching the screen, offering suggestion chips or triggering in-world events on its own.
Project Genie is adding a community gallery where users can share and remix each other's worlds. Curated starters include Living Memory, Dino Field Guide and Rock Paper Scissors, with Gemini acting as referee in the last one. Early creator builds include Villa Giallo, a room-by-room gothic mystery made with Will aka BLVCKL!GHT, and The Life of Ants, made with Anand Gandhi.
Google DeepMind shipped upgrades to the older Genie 3 in the same release. Genie 3 sessions now run up to three minutes per generation, up from 60 seconds, and worlds can be chained together for longer explorations. Users can also seed a world from a 360-degree panorama and prompt events in real time, such as summoning a dragon during a flyover.
Google DeepMind's model documentation says the same Genie 4 model can accept per-step text captions during rendering, which can be used to vary robotics simulations. All generated video and audio carry SynthID watermarks and pass through input and output safety checks.
Google DeepMind lists remaining limits as camera latency, intricate physical interaction, real-time audio precision and long-term scene permanence. The company did not publish resolution, frame-rate or benchmark figures for Genie 4.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.




