# Genie 4: Google DeepMind's world model now makes sound, dialogue and steerable worlds > The new research preview generates explorable scenes in real time with native audio and talking characters, and it is coming to Google AI Ultra's Project Genie and the Gemini API. - Publisher: NextDiff (https://nextdiff.com/) - Section: AI - Author: NextDiff Editorial - Published: 2026-10-28T06:27:42.479Z - Updated: 2026-10-28T06:27:42.479Z - Canonical URL: https://nextdiff.com/posts/genie-4-google-deepmind-world-model/ - Topics: google, gemini, ai-gaming, agents ## In short Google DeepMind announced Genie 4 on 27 October 2026, a real-time world model that turns text or an image into an explorable scene and now generates music, ambient sound and character dialogue. It is a research preview for trusted testers, with Project Genie for Google AI Ultra subscribers and Gemini API partners next. ## Key takeaways - It is Google DeepMind's first real-time world model that generates audio natively: music, ambient noise and character dialogue. - You can steer a running world with plain language, for example changing the weather, moving the camera or adding people. - It is a research preview for trusted testers now; Google AI Ultra subscribers (Project Genie) and Gemini API early-access partners are next. - DeepMind says details and object permanence still drift in long sessions, and on-screen text is unreliable. Google DeepMind's **Genie 4** turns a sentence or a single picture into a world you can walk around in, and for the first time that world makes its own sound. Announced on 27 October 2026, the [research preview](https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/) generates music, ambient noise and characters who talk back, all in real time. Until now, Genie worlds were silent. Now they are scored, voiced and steerable while you play. ![A saxophonist on a small island at dawn, generated live by the model with keyboard controls on screen](/media/c02ac0cf289a4518acf8.jpg) *Image: Google DeepMind (demo frame, press material)* ## What is Genie 4? It is a **world model**: an AI system that predicts the next moment of a scene from the current one and your input, frame after frame. You move with the keyboard (W, A, S, D and arrow keys), and the model draws what you would see next. DeepMind calls it "natively multimodal" because sound is produced in the same loop as the picture rather than added afterwards. Google says it is its first real-time world model to generate audio this way. Here is DeepMind's own launch video: https://www.youtube.com/watch?v=sFfPkUJd_Jk ## What's new compared with Genie 3? Genie 3 arrived in August 2025 and powered Project Genie, the experiment Google opened to US Google AI Ultra subscribers in January 2026. According to DeepMind and [TechTimes](https://www.techtimes.com/articles/329582/20261028/deepmind-genie-4-adds-native-audio-gemini-powered-reasoning-real-time-world-simulation.htm), the new version changes four things: | | Genie 3 (Aug 2025) | New model (Oct 2026) | |---|---|---| | Audio | none | music, ambient sound, dialogue | | Steering | text "world events" | live text edits: weather, camera, new people or objects | | Planning | SIMA agent inside the world | Gemini 3.8 Flash agent plans and checks the scene | | Look | 720p, 24 fps (reported) | "enhanced photorealism, fluid movement" (no specs given) | Google has not published the new model's resolution, frame rate or maximum session length, so treat any figure you see for those as unconfirmed. ![A stained-glass landscape world generated by the model](/media/38645c671e7d425cb868.jpg) *Image: Google DeepMind (demo frame, press material)* ## How does it work? Think of it as a video game with no game engine. Instead of a 3D level built by artists, a neural network imagines each frame on the spot, remembering enough of what it drew to keep the world consistent when you turn around. DeepMind says several Gemini models sit around that loop. Gemini Omni anchors how things look and behave in the real world. A **Gemini 3.8 Flash agent** watches the scene, breaks a goal into steps, checks progress and changes the plan when you intervene. That is what lets you type "make it rain" or "add a street musician" mid-session and have the world follow through. Characters can also hold real-time conversations, and every generated frame and sound carries Google's **SynthID** watermark. ## Who is it for? Trusted testers in film, journalism, education, gaming and engineering are using it for previsualisation, rebuilding historical scenes, prototyping games and simulation. DeepMind also showed it simulating Apptronik's Apollo humanoid robot playing checkers, the kind of synthetic training world robotics labs want. It is the same idea behind AI agents that learn by doing; our explainer on [what an LLM is](/posts/what-is-an-llm/) covers the language side of that stack. ![A humanoid robot simulated at a table](/media/e029825b2d7c4859818d.jpg) *Image: Google DeepMind (robotics demo frame, press material)* ## How can you try Genie 4? Not yet, unless you are a tester. Google says it is coming "soon" to **Google AI Ultra subscribers in Project Genie** and to early-access partners of the **Gemini API and Google AI Studio**, with a waitlist open now. There is no price or country list yet. Project Genie today is US-only inside AI Ultra, which TechTimes puts at $200 a month, so readers in India should not expect access on day one. ## Caveats DeepMind is upfront about the limits: visual details and object permanence drift in longer sessions, after many prompt edits or while you move around. Complex physics, accurate real-time audio, readable on-screen text and long-term consistency are still open problems. This is a preview, not a product. ## Questions readers ask ### What is Genie 4? Genie 4 is Google DeepMind's latest world model. It generates an interactive, explorable environment in real time from a text prompt or a reference image, and now produces sound and dialogue alongside the visuals. ### How can I use Genie 4? Right now only trusted testers can use it. Google says it is coming to Google AI Ultra subscribers through Project Genie and to early-access partners of the Gemini API and Google AI Studio, and there is a waitlist. ### What is new in Genie 4 compared with Genie 3? Google lists native audio (music, ambient sound and dialogue), more photorealism, more fluid movement and real-time steering by text. A Gemini 3.8 Flash agent also plans and checks what happens in the scene. ### Is Genie 4 free? Google has not announced a price. Project Genie currently sits inside the Google AI Ultra plan, which TechTimes reports costs $200 a month in the US. ### Is Genie 4 available in India? Google has not named countries for Genie 4. Project Genie launched in January 2026 for Google AI Ultra subscribers in the US only. ## Sources - [Google DeepMind — Genie 4: A new generation of world simulation](https://deepmind.google/blog/genie-4-a-new-generation-of-world-simulation/) - [TechTimes — DeepMind Genie 4 adds native audio and Gemini-powered reasoning](https://www.techtimes.com/articles/329582/20261028/deepmind-genie-4-adds-native-audio-gemini-powered-reasoning-real-time-world-simulation.htm) - [Quartz — Google DeepMind is unveiling Genie 4, its most realistic world model](https://qz.com/google-deepmind-genie-4-world-model-audio-102726) - [TechCrunch — I built marshmallow castles in Google's Project Genie](https://techcrunch.com/2026/01/29/i-built-marshmallow-castles-in-googles-new-ai-world-generator-project-genie)