Google’s AI worlds just learned to listen — and to talk back. On November 18, 2026, Google opened first-time access to Genie 4 inside Project Genie, its browser-based world-building prototype, for eligible Google AI Ultra subscribers. Genie 4 is Google DeepMind’s most realistic and interactive world model yet, and it changes the deal in two big ways: you can now type a new prompt in the middle of a session to rewrite the world you are standing in, and for the first time the visuals come with real-time native audio — ambient soundscapes and character dialogue that adapt as the scene unfolds. It is the same week Google has been shipping new Gemini models at a rapid clip, from the fast and cheap Gemini 4 Hydrogen to the frontier-speed Gemini 4 Neon — but Genie 4 is a different kind of model altogether. It does not answer questions. It builds a place, puts you inside it, and lets you change it while you walk.
What Is Google Genie 4?
A world model is an AI system that generates a simulated environment you can explore in real time, rather than a single image or video. Genie 4 renders walkable, interactive worlds at up to 24 frames per second, steered with keyboard controls and text prompts, and it was first announced as a research preview for trusted testers on October 27, 2026. The November 18 update is the first time it has opened to the public side of Project Genie. Project Genie itself began in January 2026 as an experimental research prototype powered by Genie 3, and it now brings together Google’s most advanced world models alongside its Gemini models — Gemini supplies reasoning and context while the Genie models render the world. Google describes Genie 4 as its highest-fidelity option for visuals, physics and native audio. Genie 3 stays available in the model picker for longer runs or worlds seeded from 360-degree panorama images.
Rewrite the World Mid-Stroll

The headline feature is live editing. With Genie 4, you no longer have to start over when a world is not quite right. Type a new prompt while you navigate and the environment changes around you in real time: shift the weather, drop in a character, or redirect the action. That is a sharp break from Genie 3’s original one-minute generations, where a world was essentially a fixed clip you walked through. The practical effect is closer to directing than prompting — you explore, you notice what is missing, and you ask for it on the spot. Google is positioning this as the core creative loop of Project Genie: generate a world, step inside it, shape it, and keep shaping it until it matches the vision in your head.
Worlds That Talk Back: Native Audio Arrives
The second big change is sound. For the first time in the Genie line, the model generates real-time native audio in the same pass as the visuals — ambient soundscapes that shift with the scene, music that fits the mood, and characters who speak dialogue that reacts to what you do. Earlier world models were silent by default, which made even the prettiest generated landscape feel like a diorama. Native audio is what makes a place feel inhabited, and it is the feature that most clearly separates Genie 4 from Genie 3. Google says the audio is generated alongside the visuals rather than layered on afterwards, so a storm you summon mid-session should sound like a storm as well as look like one. That said, Google itself lists real-time audio precision among the model’s unresolved challenges, so expect the soundtrack to be impressive but not yet flawless.
Genie 3 Gets Longer Runs, Too
Genie 4 is the headline, but the November 18 update also upgrades the older model. Genie 3 sessions now run for up to 3 minutes per generation, up from 60 seconds, and you can link multiple worlds together for longer journeys. You can also seed new worlds from 360-degree panorama images, and trigger real-time in-world events with natural-language prompts while you explore. The message is clear: Genie 4 is the fidelity pick, Genie 3 is the endurance pick, and both are now more flexible than they were at launch. For creators, the longer sessions matter as much as the new model — a one-minute world is a demo, but a three-minute world chained to another is the beginning of a story.
The Project Genie Agent and the New Gallery

The update also adds a conversational Project Genie agent that takes creation beyond a single text or image prompt. Describe your vision in natural language and the agent orchestrates Google’s suite of multimodal models to build your world — establishing its signature look and pairing it with a custom background score tailored to the mood. While you explore, Gemini models can watch what is happening on screen, surface context-aware suggestion chips for where to go next, or trigger in-world events automatically as your journey unfolds. There is also a new community gallery for discovering and remixing worlds: turn personal photos into explorable spaces in “Living Memory,” wander prehistoric habitats in “Dino Field Guide,” or play “Rock Paper Scissors” with Gemini as a real-time referee. Google also showcased collaborations with creators who had early access — “Villa Giallo,” a gothic manor mystery built with creative technologist Will aka BLVCKL!GHT, and “The Life of Ants,” a journey through an ant colony from an ant’s perspective built with filmmaker Anand Gandhi. Once you have crafted something, you can publish it to the gallery for others to explore and remix.
Who Can Try It — and What It Costs
Access is gated. The new Project Genie experience is rolling out to eligible Google AI Ultra subscribers starting today, and Google AI Pro members can only join a waitlist on the same page. Google’s announcement names no price for Genie 4 itself — it is bundled into the Ultra subscription — and there is no free tier. That puts Genie 4 in the same premium tier as the rest of Google’s frontier experiments: available to paying subscribers first, with the wider public waiting in line. If you are an Ultra subscriber, the entry point is the Project Genie page on Google Labs, which runs entirely in the browser with no install. If you are on AI Pro, the waitlist is the only door for now.
Still a Research Prototype: The Honest Limits
Google is careful to say that Genie 3 and Genie 4 remain world models in the frontier research space, and the November 18 post lists the challenges it is still working on: camera latency, intricate physical interactions, real-time audio precision, and long-term scene permanence — the tendency for objects to change when you look away and come back. Those are real limits on how far you can push a world today. On the safety side, all video and audio generated in Project Genie carries SynthID, Google’s imperceptible watermark for identifying AI-generated content, and all model inputs and outputs go through safety screenings. The honest read: this is a research preview with a subscription gate, not a finished product. The worlds are more realistic, more interactive and now audible — but they are still short-lived, occasionally inconsistent, and available only to a small slice of users.
The Bottom Line
Genie 4 is the clearest sign yet of where Google thinks generative AI is heading: from generating content to generating places. Mid-session editing turns a world from a fixed output into a conversation, native audio turns it from a picture into a place, and the agent and gallery turn it from a solo experiment into something shareable. The same week has brought GPT-6.2 Astra cutting prices by 20% and a flood of new Gemini capabilities, but Genie 4 is the one that feels most like a glimpse of a different medium — not a smarter chatbot, but a world you can walk into and rewrite as you go. The limits are real and Google names them plainly. The direction of travel is just as plain.
Sources
Google, “Genie 4 is now available,” The Keyword, November 18, 2026: https://blog.google/innovation-and-ai/models-and-research/google-deepmind/genie-4-now-available-project-genie/
Project Genie, Google Labs: https://labs.google/projectgenie
Unite.AI, “Google Brings Genie 4 to Project Genie for Eligible AI Ultra Users,” November 18, 2026: https://www.unite.ai/google-brings-genie-4-to-project-genie-for-eligible-ai-ultra-users/
Related reading: Gemini Decisions API and Firefly.

