Type a sentence, and a world appears. A rain-slick street in Tokyo, neon signs buzzing overhead, steam rising from a food cart on the corner. You step forward, and the street responds.
The rain keeps falling. A shopkeeper turns toward you and speaks. Now type another sentence, and the rain stops. The sun comes out. The shopkeeper waves you inside.
That is the promise of Google Genie 4, and as of November 18, it is no longer locked inside Google’s labs. Eligible Google AI Ultra subscribers can now walk through these living worlds themselves, straight from a browser.
Google DeepMind announced Genie 4 on October 27 as a research preview for trusted testers only. Three weeks later, the model has graduated to its first consumer audience.
The rollout arrives inside a refreshed Project Genie experience, Google’s browser-based playground for interactive worlds. It brings more than just the new model: a creation agent, a community gallery, and upgrades to the older Genie 3 model all arrive in the same update, according to Google.
What Google Genie 4 actually does differently
World models are a different kind of AI. A chatbot answers in words. An image model draws a still picture. A world model draws the next moment, frame by frame, as you move through a scene.
There is no game engine underneath and no stored 3D map. The world exists only because the model keeps imagining it, one frame at a time, guided by your keyboard and your words.
Genie 4 is the most realistic and interactive world model Google has built, and its headline trick is that the world no longer waits for you to start over. You can prompt mid-session, typing a new instruction while you are already inside the scene.
Change the weather. Drop in a character. Redirect the action entirely. The world reshapes itself around you without breaking the spell.
The second leap is sound. For the first time in the Genie line, the model generates native audio in the same pass as the visuals. Ambient soundscapes swell and fade as the scene changes.
Music follows the mood. Characters speak, and their dialogue reacts to what is happening around them. A silent diorama becomes a place with a soundtrack, and that changes how alive it feels.
The Project Genie agent builds worlds with you
Until now, making a world in Project Genie started with a single prompt or an image. The new Project Genie agent turns that into a conversation.
Describe your vision in plain language, and the agent orchestrates Google’s suite of multimodal models to build the opening scene. It chooses the visual style and composes a custom background score to match the mood you described.
The agent does not leave once the world is built. Gemini models can join the session as you explore, watching what is happening on screen and offering suggestion chips for where to go next.
Sometimes they go further, triggering in-world events on their own as your journey unfolds. The line between player and director gets very thin, very fast.
A gallery of worlds to explore and remix
Not everyone wants to build from a blank page, so Google has added a community gallery to Project Genie. It mixes curated worlds with creations from the community, and every one of them can be explored or remixed into something of your own.
Three of the featured experiences show off what the technology can do. “Living Memory” turns personal photos into explorable spaces, letting you step back inside a captured moment.
“Dino Field Guide” wanders through prehistoric habitats grounded in 360-degree panoramas. And “Rock Paper Scissors” pits themed objects against each other on a desk, with Gemini acting as a real-time referee.
The gallery also features collaborations with creators who had early access. Creative technologist Will, known as BLVCKL!GHT, built “Villa Giallo,” an atmospheric mystery that plays out room by room inside a gothic manor.
Filmmaker Anand Gandhi created “The Life of Ants,” which invites you to navigate a complex colony from an ant’s perspective, drawing on Genie 4’s photorealism. Once you have crafted something you like, you can publish it to the gallery for others to explore and remix.
Genie 3 gets longer sessions and panorama worlds
Genie 4 is the star of the update, but Genie 3 is not being retired. It stays in Project Genie’s model picker for longer runs and panorama-seeded spaces, and it picks up several upgrades of its own.
Sessions now run for up to three minutes per generation, up from 60 seconds, and multiple worlds can be linked together for longer journeys.
You can also seed new worlds from 360-degree panorama images, and real-time in-world events now work in Genie 3 too. That means you can change the environment with natural-language prompts while you explore.
For anyone who wants a longer, steadier stroll rather than the sharpest visuals, the older model still has a role.
Who can try Google Genie 4, and what it costs
Access is the catch. Google Genie 4 is rolling out to eligible Google AI Ultra subscribers, the company’s top-tier plan, which costs $99.99 a month in the United States.
Ultra carries five times the usage limits of the cheaper AI Pro plan. Project Genie runs entirely in the browser at labs.google/projectgenie, so there is nothing to install, but you need to be 18 or older and signed in with a Google account.
Google AI Pro members are not shut out entirely. They can join a waitlist on the same Project Genie page, and Google says the experience is available in more than 140 countries.
Paying for Ultra does not guarantee instant access either, since Google describes the rollout as going to eligible subscribers, with availability opening up over time rather than all at once.
There is still no word on when Genie 4 will reach the Gemini API or Google AI Studio, and Google has announced no general availability date beyond this rollout. For now, Project Genie is the only door in.
Why world models are Google’s next big bet
Genie 4 arrives at the end of a remarkable month for Google DeepMind. The lab has shipped new Gemini models at a relentless pace, including Gemini 4 Neon just yesterday.
It has also pushed AI into robots with Gemini Robotics 2.5 on Mercedes-Benz factory floors. World models are the next frontier in that same push: AI that does not just describe reality, but simulates it.
The uses go well beyond entertainment. Game designers could prototype levels by describing them. Filmmakers could scout locations that do not exist.
Robotics teams could train machines in endless simulated environments before a single real-world test. A world model that can be edited in plain language is, in effect, a rehearsal space for anything that moves through space and time.
That is also why the race is heating up. Every major lab is now chasing interactive simulation, and the company that makes worlds feel real, responsive, and cheap to generate will shape how the next generation of games, films, and training simulations get made.
Google clearly wants that company to be Google.
The honest limits: it is still a research prototype
Google is refreshingly open about what Genie 4 cannot yet do. The company still calls both Genie models world models in the frontier research space, and its own announcement lists the open problems.
Camera latency, intricate physical interactions, real-time audio precision, and long-term scene permanence are all still being worked on. In plain terms, objects can change when you look away, physics can wobble, and long sessions can drift from where they started.
There are guardrails too. All video and audio generated in Project Genie carries SynthID watermarking, and all model inputs and outputs pass through safety screenings.
Google has also published a Genie 4 model research overview for anyone who wants the technical detail behind the promises.
None of that dims the moment. Three weeks ago, Genie 4 was a research preview for a handful of trusted testers. Today, anyone with an AI Ultra subscription can type a sentence and step inside the world it creates.
The worlds are imperfect, the access is narrow, and the price is steep. But the door is open, and for the first time, the public can walk through it.



