Google DeepMind has opened Genie 4, its interactive world model, to eligible Google AI Ultra subscribers through Project Genie, and users can now change a world by typing while they explore it. Google announced the rollout on November 18 in a post on its blog written by Diego Rivas, a group product manager at Google DeepMind, and Pavan Kumar, a group product manager at Google Labs.
The change that matters for builders is the editing loop. In earlier versions you described a world, then stepped into it. With Genie 4 you type a new prompt mid-session, and the environment shifts in real time. Google’s examples are changing the weather, dropping in a character, or redirecting the action. Genie 4 also generates audio natively, and Google says the sound shifts with the action, covering background ambience and spoken lines for characters.
Genie 3 does not go away. Google keeps it in the model picker for people who prefer extended runs or spaces built from panoramas. It also received upgrades: a single generation can now run up to 3 minutes, up from 60 seconds, and several worlds can be chained for longer journeys. Genie 3 can also start a world from a 360-degree panorama image, and it takes natural-language prompts for in-world events while you explore.
The release adds a Project Genie agent. Rather than one prompt, you talk through what you want, and the agent coordinates Google’s multimodal models to produce the opening visuals and a custom background score. During a session, Gemini models can watch the screen, propose clickable next destinations, or trigger events on their own.
A community gallery rounds out the package. It mixes curated worlds with user creations, and anyone can remix an entry by changing its structure, soundtrack, or rules, then publish the result. Examples include “Living Memory,” which turns personal photos into explorable spaces, and “Rock Paper Scissors,” where Gemini referees a match between themed objects on a desk. Google also built “Villa Giallo,” a room-by-room mystery in a gothic manor, in collaboration with the creative technologist Will, who goes by BLVCKL!GHT, and “The Life of Ants,” a colony seen from an ant’s perspective, with filmmaker Anand Gandhi.
Access is narrow. Eligible Google AI Ultra subscribers can use the new experience starting November 18, while Google AI Pro members can join a waitlist on the same page. Google has not said when the waitlist opens up, and the announcement gives no pricing or usage limits for the new features.
The company is open about the limits of the technology. It describes both Genie 3 and Genie 4 as frontier research, and names four open problems: camera latency, complex physical interactions, how precisely audio tracks events, and whether a scene stays consistent over time. All video and audio from Project Genie carries SynthID watermarking, and Google runs safety screenings on what goes into and comes out of the models. The post contains no independent evaluation of how well the model handles those problems, and its claims about realism come from Google itself.
The gap between a demo and a tool shows up in those unsolved list. A world that forgets what you built a minute ago, or lags when you turn the camera, is fine for a gallery piece and not yet usable as a prototype for a game or a simulation. Teams exploring world models for interactive media should treat Genie 4 as a way to test prompt-driven direction and audio, and hold off on any production plan until scene permanence and latency improve.
Google DeepMind, Genie 4 is now available, published on the Google blog on November 18, 2026.