Google DeepMind has rolled out Genie 4 to eligible Google AI Ultra subscribers, transforming Project Genie from a research prototype into a consumer product for generating and exploring interactive AI-generated worlds. The update brings first-time public access to Genie 4, a conversational build agent, and a community gallery where users can share and remix their creations, Google announced.

Project Genie was introduced earlier this year as an experimental research prototype, and early testers told Google they wanted deeper control over what happens inside generated worlds and easier ways to build and share them. The new release, covered widely by the developer community since it topped discussion forums this week, addresses both requests. For ongoing coverage of the latest AI developments from Google DeepMind and every other major lab, AI Buzz Wire follows each rollout as it happens.

Edit Any World Mid-Session

Genie 4 is Google DeepMind's most realistic and interactive world model yet, and its defining feature is live editing: users can prompt mid-session to change the world they are navigating in real time. "Shift the weather, drop in a character, or redirect the action" with a typed prompt, Google said in its announcement.

For the first time, Genie 4 pairs those visuals with real-time native audio that adapts as the scene unfolds, ranging from ambient soundscapes to character dialogue. Google demonstrated turning a natural language prompt into an interactive spacewalk, then summoning an alien spaceship on command, and remixing a hillside cafe into a glass pavilion in the desert.

A Conversational Agent Builds Your World

Creating a world no longer requires a single perfect prompt. The new Project Genie agent orchestrates Google's suite of multimodal models through ordinary conversation: users describe their vision, and the agent crafts initial visuals establishing the world's signature look, paired with custom background scores tailored to the mood.

Gemini models can also join the session as it runs, observing what happens on screen, surfacing context-aware suggestion chips for where to go next, or triggering in-world events automatically as the journey unfolds. The result, Google said, makes crafting an interactive universe "as effortless as having a conversation."

A Gallery Built for Remixing

The new community gallery lets Ultra subscribers publish custom worlds for others to explore and remix. Google's own curated examples show the range: "Living Memory" transforms personal photos into explorable spaces, "Dino Field Guide" wanders through prehistoric habitats grounded in 360-degree panoramas, and "Rock Paper Scissors" pits themed objects against one another on a desk with Gemini serving as a real-time referee.

Genie 3 Remains in the Model Picker

Despite Genie 4's debut, Genie 3 stays available in Project Genie's model picker for cases where it remains the better tool — longer runs and panorama-seeded spaces. Google highlighted Genie 3 updates including real-time promptable events that connect different worlds into one continuous flight across forests, underwater ruins and icy seas, and 360-degree panorama grounding that kept Petra's canyon and facade consistent as the camera moved.

In one demonstration, Genie 3 combined longer runs, panorama grounding and promptable events to transform a world mid-flight — even summoning a rideable dragon on command.

Access and Availability

Genie 4 is rolling out now to eligible Google AI Ultra subscribers, who can build, share and remix worlds in the community gallery today. Google AI Pro members can join a waitlist. Everything runs directly in the browser through the Project Genie website, with no downloads required.

From Prototype to Product in Three Weeks

The speed of this rollout is notable. Project Genie existed as a limited research experiment only weeks ago, and Google says the update is a direct response to what early testers asked for: more control, and easier sharing. Rather than a slow research-to-product pipeline, DeepMind shipped consumer access, a creation agent and social infrastructure in one step — a sign of how aggressively Google is pushing its world-model program toward the mainstream.

It also reflects a broader shift in how the company launches AI products. Instead of waiting for a polished flagship, Google is iterating in public, using its subscriber base as a live testing ground while keeping an explicit research lineage — Genie 3 — available for the workloads the new model does not yet handle best.

Why World Models Matter

World models — AI systems that simulate interactive environments rather than merely predicting text or rendering static frames — are widely seen as a potential foundation for gaming, robotics training and embodied AI research. With Genie 4, Google is the first major lab to put a frontier world model in the hands of general consumers at scale, complete with social features that turn world-building into shareable content.

The rollout also intensifies the competitive race in generative media. As world models, video generators and game engines converge, the line between playing a game and prompting one is beginning to blur — and Google now has a consumer-facing product to test exactly where that line sits.

Stay Ahead of AI

From world models to frontier LLMs, the AI landscape is shifting weekly. Track the latest AI developments at AI Buzz Wire.

Read more AI news →