Genie 4: what Google DeepMind's new world model is and what it means for creators
On Tuesday, October 27, 2026 Google DeepMind unveiled Genie 4, a world model that generates explorable worlds in real time, complete with sound, music and characters who talk. While you move through the scene you can type what should change: the weather, the camera, a new character. For now only selected testers can use it. Here is what it does, when it arrives and why it matters if you make video.
Leggi questo articolo in italiano
What Genie 4 is and when it was announced
Genie 4 is Google DeepMind's new world model, announced on October 27, 2026: an AI model that generates an explorable world in real time, with images, sound and characters that react to what you do. Unlike a video generator such as Veo or Kling, which produces a finished clip to watch, a world model creates a scene you move through: every step or command changes what happens next.
The official Google DeepMind post, written by Raia Hadsell, Diego Rivas and Hang Qi, calls it the company's most realistic and interactive world model yet. Genie 4 builds on the Gemini family:
- Gemini Omni anchors the simulation in real-world appearance and behavior;
- Gemini 3.8 Flash acts as a guiding agent: it watches the scene, breaks a goal into steps, checks that each step succeeds and replans when the world shifts or you step in;
- other Gemini models help keep the world coherent and continuous.
Its predecessor, Genie 3, was introduced by Google DeepMind in August 2025, also as a research preview.
What it does: native audio, live prompts, better physics
Genie 4's three main changes are audio generated together with the visuals, the ability to change the scene with a prompt while you explore it, and more realistic physics for objects and people.
- Native audio. Google DeepMind says Genie 4 is its first real-time world model to generate audio natively: wind, waves and ambient noise that fit the place, plus music and character dialogue that adapt to what happens.
- Live direction in plain words. You move with keyboard and mouse and, whenever you want, type what should change: the weather, the camera position, one more person or object. Gemini adapts your prompt to the scene and keeps the world moving between commands.
- Physics and people. The model simulates scenarios such as aerodynamics and collisions; characters move naturally, react to their surroundings and can be steered in real time.
On duration, the sources word it differently. The model page speaks of “a few minutes of continuous interaction” rather than extended sessions; TechCrunch reports that Genie 4 keeps a scene consistent for up to two minutes, double Genie 3's limit. Resolution and frame rate have not been disclosed.
Who can use it, when, and what it costs
As of October 28, 2026 Genie 4 is only available as a research preview for selected testers: you cannot try it right away, but you can join the Project Genie waitlist. Project Genie is Google's experimental tool for generating interactive worlds from text or images.
| Who | How | When |
|---|---|---|
| Selected testers | Research preview | From October 27, 2026 |
| Google AI Ultra subscribers | Project Genie | “Coming soon”; in the coming weeks according to TechCrunch |
| Partner developers | Early access via Gemini API and Google AI Studio | “Coming soon” |
Google has announced neither a specific price nor the countries where it will be available. On safety, Google DeepMind says it ran internal evaluations before the preview and that all video and audio produced by Genie 4 will include SynthID, its invisible watermark, in both picture and sound.
The limits Google itself admits
Google DeepMind calls Genie 4 an “early, experimental model” and lists its own limits: details can drift over time, in-world text is often unreadable and audio is not always accurate.
- Object permanence. Visual details and objects can drift over time, especially in longer sessions, after many prompt edits or while you move through the scene.
- Text. Writing in the world is often legible only if it already appears in the prompt.
- Real places. Existing locations are not reproduced with perfect accuracy.
- Several characters at once. Multi-agent interactions remain an open challenge.
- Audio. Real-time audio accuracy is still being improved.
Diego Rivas, a Google DeepMind engineer, told TechCrunch it is still early days and the full product experience is not settled. Tejas Kulkarni, research director, sees the future in “hybrid world models,” where Gemini builds structure and narrative and Genie simulates picture and sound. Genie 4 is not alone: TechCrunch notes that Runway launched GWM Worlds 2, a playable world model with audio, a couple of months earlier.
What it means for you if you make video or content
For video makers, Genie 4 mainly promises one thing: testing a scene idea by moving inside it, instead of generating one clip at a time. Google DeepMind names a “draft mode” for previsualization, with several “takes” of the same scene, as a first creative use; the testers quoted by TechCrunch point to ideation and storyboarding for film.
- Do not change your workflow today. Genie 4 is not accessible yet: finished clips still come from video generators. Read what changes with Kling 4.0 and our Google Flow guide.
- Think in scenes, not shots. Get used to describing place, light, sound and characters as a world: that is how prompts for world models will be written.
- Join the waitlist if you have AI Ultra. Project Genie will be the first way in for subscribers.
- Keep an eye on Gemini Omni. It is the model that anchors Genie 4 to reality; see our Gemini Omni guide and our article on Gemini 4 Argon.
The skill that matters stays the same: directing. AI Filmaking teaches you to write scenes, shots and camera moves for AI video, skills you will also need with world models; for a quick start there is Create stunning AI videos in 5 minutes, and if you want the video done for you there is the 30-second AI video service. If AI-made games interest you, also read about Higgsfield Playable. For a complete method there is The AI Creator System, and here is how we use artificial intelligence.
Frequently asked questions
What is Genie 4?
It is Google DeepMind's world model announced on October 27, 2026: it generates interactive worlds in real time that you explore with keyboard and mouse, with native audio, and you can change them by typing a prompt as the scene runs.
Can I use Genie 4 now?
No, unless you are one of the selected testers. Google will bring it soon to Google AI Ultra subscribers in Project Genie, which has a waitlist, and to early-access partners of the Gemini API and Google AI Studio.
How much does Genie 4 cost?
Google has not announced a price or the countries where it will be available. The first consumer access will come through the Google AI Ultra subscription.
What is the difference between a world model and a video generator?
A video generator such as Veo or Kling creates a finished clip to watch. A world model such as Genie 4 creates a scene you move through, which reacts in real time to your controls and prompts.
Read next


