Google opens Genie 4 Project Genie to Ultra members with world-building agent and gallery
By GameSlash
November 18 rollout gives eligible Ultra subscribers interactive world editing, a creation agent and remix gallery; Genie 3 separately adds three-minute runs.
Google announced on November 18, 2026 that eligible Google AI Ultra subscribers can now use a new Project Genie experience powered by Genie 4. This is an access launch, not the original unveiling of the research model in October: it also adds a world-building AI agent and a community gallery of interactive experiences to explore and remix. Official announcement.

What changes with Genie 4 access
In an explorable world, a subscriber can type instructions mid-session to alter weather, place a character or redirect events as the simulation runs. Genie 4 also generates native environmental audio, music and reactive dialogue. Its model card specifies real-time 720p output at up to 24 frames per second. These are world-model simulations, not exported Unity, Godot or Unreal levels.

AI agent and community gallery
The new Project Genie agent helps plan a world in conversation, producing initial visual concepts and background scores. Gemini can observe a session to propose next actions or initiate events. Subscribers can explore and remix publicly shared worlds in a gallery, including experimental interactive experiences. This matters to game creators who want fast mood and interaction references without mistaking a simulated world for production-ready gameplay code.

Genie 3 also changed, but it is a separate model
Google also extended its selectable Genie 3 sessions from 60 seconds to up to three minutes per generation, added links between generated worlds and 360-degree panorama-based creation. The three-minute feature applies to Genie 3, not a stated Genie 4 runtime. Google AI Pro users can join a waitlist; availability for Ultra subscribers still depends on eligibility.
A practical game previsualization test, and limitations
A development team could describe a room or level mood, enter the world, request a weather change and compare candidate atmosphere or camera angles before rebuilding the idea in an engine. Google has not announced a production-ready mesh, collision, scripting or engine-project export flow. The model card warns that small details, physical interactions and object placement can drift during long or complex runs, so human validation remains essential.
This November access story is separate from GameSlash's October Genie 4 research-preview story. References: Google's new availability announcement and Genie 4 model card.