Google DeepMind made Genie 4 available to eligible Google AI Ultra subscribers through Project Genie on 18 November, according to its release announcement. The browser-based experience lets subscribers generate and explore interactive worlds, then change them with prompts while moving through them. It also adds a creation agent and a gallery of worlds that others can explore and remix.
Key points
- Google says Genie 4 can change an environment during a session and generate audio alongside its visuals.
- Genie 3 now supports sessions of up to 3 minutes per generation and worlds seeded from 360-degree panorama images.
- A Project Genie agent helps create worlds from descriptions, and a gallery lets users explore and remix them.
- Access is available to eligible Google AI Ultra subscribers; Google AI Pro members can join a waitlist.
Genie 4 changes worlds during exploration
The subscriber release follows a narrower introduction. Auganix, a publication covering augmented and virtual reality, reported on 11 November that Genie 4 was in research preview for trusted testers, with a Project Genie waitlist open. An earlier AI Affairs report covered its unveiling with audio and two-minute simulations as a research preview.
Project Genie began earlier in 2026 as an experimental way to generate and explore interactive worlds. Google says the new experience brings its world models together with Gemini models, allowing a person to describe or supply an image for a world and then move through the result in a browser. A world model generates the scene as exploration unfolds: unlike a fixed video, the environment has to respond to the direction taken and to changes requested along the way.
In Genie 4, those requests can arrive after exploration has started. Google says a prompt entered during a session can alter the weather, introduce a character or redirect the action without beginning a new world. It also says Genie 4 generates audio as the scene unfolds, including background sounds and character dialogue. The sound is part of the model’s output, adapting to events rather than merely accompanying an already finished sequence of images.
The company describes camera latency, complicated physical interactions, precise real-time audio and keeping scenes consistent over long periods as continuing research problems. In its account of the earlier preview, Auganix reported Google’s warning that visual details and the persistence of objects could drift during longer sessions or after several complex edits.
Genie 3 sessions reach 3 minutes
Genie 3 remains selectable within Project Genie. Google says a generation can now run for up to 3 minutes, compared with 60 seconds before this update, and that separate worlds can be linked for longer journeys. It also supports changes prompted during exploration, alongside worlds seeded from 360-degree panorama images.
A panorama supplies views in every direction from a point, giving a generated world a visual reference as the viewpoint turns. Google demonstrates the feature with a panorama of Petra, where it says the canyon and facade remain consistent as the viewer looks around.
Google positions the two models for different choices within the same experience: Genie 4 for the visuals, physics and audio it describes as its highest-fidelity output, and Genie 3 for longer runs or panorama-seeded spaces. Google says Genie 3 remains in Project Genie’s model picker for longer runs or panorama-seeded spaces.
Project Genie adds an agent and gallery
The new Project Genie agent turns a natural-language description into a starting world by coordinating Google’s multimodal models. According to Google, it prepares initial visuals and a background score suited to the description. Gemini models can also observe the screen during exploration, offer suggestions for where to go next or trigger events inside the world. For people building experiences on Project Genie, that places creation, changes during a session and sharing in one interface.
The gallery combines curated worlds with those published by the community. Google says visitors can enter a world, alter its structure, soundtrack or rules with the agent, and publish their own version for others to explore and remix. Among its examples, “Living Memory” turns personal photos into explorable spaces, while “Dino Field Guide” uses 360-degree panoramas as a starting point for prehistoric habitats.
Google also describes gallery worlds made with early collaborators. “Villa Giallo”, created with Will, also known as BLVCKL!GHT, is a mystery set inside a manor. “The Life of Ants”, built with filmmaker Anand Gandhi, lets visitors navigate a colony from an ant’s perspective.
Google says every video and audio output generated in Project Genie carries SynthID watermarking and that inputs and outputs pass through safety screening. Eligible Google AI Ultra subscribers can access the new experience from 18 November, and Google AI Pro members can join the Project Genie waitlist.