NewsStocksGoogle DeepMind Makes Genie 4 World Model Available to Google AI Ultra Subscribers

Google DeepMind Makes Genie 4 World Model Available to Google AI Ultra Subscribers

Author: Google DeepMind Blog·

Key Takeaways

  • •Google DeepMind launched Genie 4, which the company calls its most realistic and interactive world model yet, to eligible Google AI Ultra subscribers on November 18, 2026, while Google AI Pro members can join a waitlist.
  • •Genie 4 allows users to edit environments in real time with natural-language prompts mid-session and, for the first time in the Genie line, pairs visuals with adaptive native audio ranging from ambient soundscapes to character dialogue.
  • •Genie 3 received upgrades that extend exploration sessions from 60 seconds to three minutes, enable linking multiple worlds into longer journeys, and allow new worlds to be seeded from 360-degree panorama images.
  • •A conversational Project Genie agent lets users build custom worlds by describing their vision, while Gemini models can join sessions to surface navigation suggestions and trigger in-world events automatically.
  • •All video and audio generated in Project Genie carries SynthID watermarking and passes safety screenings, and Google DeepMind identified camera latency, intricate physical interactions, real-time audio precision, and long-term scene permanence as ongoing research challenges.
Google DeepMind Makes Genie 4 World Model Available to Google AI Ultra Subscribers

Google DeepMind announced on November 18, 2026 that Genie 4, which the company calls its most realistic and interactive world model yet, is rolling out to eligible Google AI Ultra subscribers through an updated Project Genie experience. The release pairs first-time access to Genie 4 with new upgrades to the Genie 3 model, a conversational Project Genie agent for building custom worlds, and a community gallery where users can discover, explore, and remix generated worlds.

Writing in a blog post, Diego Rivas, Group Product Manager at Google DeepMind, and Pavan Kumar, Group Product Manager at Google Labs, said the rollout offers "more responsive, social, and interactive world generation." Users can now edit environments in real time with natural language, add native audio, and share their creations in the new gallery.

Genie is Google DeepMind's family of world models — AI systems that generate playable, interactive environments. Unlike conventional text, image, or video generators, world models are designed to simulate environments that users can move through and interact with in real time. The line began with Genie 1 in February 2024, followed by Genie 2 in December 2024 and Genie 3 in August 2025, with Project Genie serving as the browser-based layer that lets users generate, step inside, and shape worlds built by the models. Four generations have arrived in under three years, and this rollout puts the newest one in users' hands for the first time — moving world simulation from a research demo into a consumer subscription.

From Research Prototype to Subscriber Rollout

Google DeepMind introduced Project Genie earlier this year as an experimental research prototype that lets users generate and explore interactive worlds. According to the post, early testers told the team they wanted deeper control over what happens inside those worlds, as well as easier ways to build and share them. The company said the new experience makes these worlds "far more responsive, creative, and social," and brings together its most advanced world models alongside its Gemini models so users can "generate, step inside, and shape worlds directly from your browser."

Alongside first-time access to Genie 4, the update includes new upgrades to Genie 3, the Project Genie agent that helps users build their vision, and a community gallery for discovering, exploring, and remixing worlds.

Genie 4: Real-Time Editing and Native Audio

With Genie 4, prompting no longer ends at world creation. Users can type a new prompt mid-session while navigating a world and edit the environment in real time: shifting the weather, dropping in a character, or redirecting the action. For the first time, the model also pairs its visuals with real-time native audio that adapts as a scene unfolds, ranging from ambient soundscapes to character dialogue. Google positioned Genie 4 as its highest-fidelity option, delivering its strongest visuals, physics, and native audio.

Genie 3: Longer Sessions, Panorama Grounding, World Chaining

Genie 3 remains available in Project Genie's model picker for cases where users want longer runs or panorama-seeded spaces. The updates announced on November 18 include:

  • Extended sessions and world chaining: Exploration can now run up to 3 minutes per generation, up from 60 seconds — tripling the previous cap — and users can link multiple worlds together for longer journeys.
  • 360-degree panorama image grounding: New worlds can be seeded from 360-degree panorama images.
  • Real-time in-world events: Environments can be changed dynamically with natural-language prompts while exploring.

Google demonstrated how Genie 3's real-time promptable events can connect separate worlds into a single continuous flight across forests, underwater ruins, and icy seas — transforming the environment mid-flight and even summoning a dragon for the user to ride on command. In a separate showcase, 360-degree panorama grounding kept the canyon and facade of Petra, the UNESCO World Heritage archaeological site in Jordan, consistent as a viewer looked around.

Building Worlds Through Conversation

The Project Genie agent extends creation beyond a single text or image prompt. Users describe a vision in natural language, and the agent orchestrates Google's suite of multimodal models to produce initial visuals that establish a world's signature look, paired with custom background scores tailored to the mood. Google said crafting an interactive universe is now "as effortless as having a conversation."

Gemini models can also be brought into a session while exploring. The company said they observe what is happening on screen, surface context-aware suggestion chips for where to go next, or trigger in-world events automatically as a journey unfolds.

Examples highlighted in the announcement include turning a natural-language prompt into an interactive spacewalk in which an alien spaceship appears on command, and remixing a hillside cafe into a glass pavilion in the desert.

Gallery: Curated Worlds, Creator Collaborations, and Remixing

The new gallery gathers curated worlds alongside creative work from the community, letting users jump straight into exploring. Featured experiences include "Living Memory," which transforms personal photos into explorable spaces; "Dino Field Guide," which lets users wander through rich prehistoric habitats grounded in 360-degree panoramas; and "Rock Paper Scissors," which pits themed objects against one another on a desk with Gemini serving as a real-time referee.

The gallery also showcases collaborations with creators who had early access to the tools. With creative technologist Will aka BLVCKL!GHT, Google DeepMind created "Villa Giallo," an atmospheric, room-by-room mystery inside a gothic manor built with Genie 4. With filmmaker Anand Gandhi, the team built "The Life of Ants," which draws on Genie 4's photorealism to let users navigate a complex colony from an ant's perspective.

Beyond browsing curated content, users can launch their own ideas from a simple prompt or remix existing gallery experiences with the Project Genie agent — tweaking the underlying world structure, soundtrack, or rules — and then publish their creations to the community gallery for others to explore and remix.

Watermarking, Safety Screens, and Open Challenges

Google DeepMind said it is committed to developing and rolling out interactive world models responsibly. All video and audio generated in Project Genie carries SynthID watermarking, the company's watermarking technology for AI-generated content, and all model inputs and outputs go through safety screenings.

The company emphasized that Genie 3 and Genie 4 remain world models in the frontier research space. As the research evolves, it is actively tackling key challenges including camera latency, intricate physical interactions, real-time audio precision, and long-term scene permanence — a list that amounts to a public checklist of what to watch as the models mature beyond today's three-minute sessions. A deeper look at model limitations and progress is available in the Genie 4 model research overview.

Availability

Eligible Google AI Ultra subscribers can try the new Project Genie experience starting today, and Google AI Pro members can join the waitlist on the same page. Google directed users to the Project Genie website to start building their own worlds. The post did not specify a timeline for broader access beyond the waitlist, leaving the pace of any wider rollout the next signal to watch.