Google DeepMind introduced Genie 4: worlds you walk through and change by typing
Direct answer
On October 27, 2026, Google DeepMind announced Genie 4, its newest world model: from text or an image it generates a scene you can walk through in real time with a keyboard, with native audio including music, ambient sound and character dialogue. While the scene runs, you can type a change, such as rain, a different camera angle or a new person, and the world adjusts. For now it is a research preview for invited testers, with access coming soon to Google AI Ultra subscribers through Project Genie and early-access partners of the Gemini API and Google AI Studio. A waitlist is open.
What changed from Genie 3
- Native audio for the first time in the line: sound is generated together with the picture, not added afterwards. Characters converse and react to what happens in the scene in real time.
- Directed by Gemini: Gemini Omni supplies appearance and movement, and Gemini 3.8 Flash acts as director: it splits a goal into steps, checks each frame and replans when the user intervenes, per The Verge's coverage.
- Coherent scenes for longer: TechCrunch reports a scene now holds together for up to two minutes, double Genie 3's limit of about one minute.
- Better physics and characters: Google describes better physics, collisions and aerodynamics, and characters that react more naturally. DeepMind itself notes limits: objects still drift over time, text inside the scene is not always legible, and long-term consistency remains unsolved.
The word 'first' deserves an asterisk
- Google's claim: DeepMind presents Genie 4 as its first real-time world model that generates audio natively.
- The nuance: specialist coverage notes that Runway released GWM Worlds 2 a couple of months earlier, also playable and also with audio. The word 'first' is, for now, Google's claim, not an established fact.
- Watermark: Google says all generated video and audio will carry SynthID, its hidden watermark for AI-made content. For a company, that detail matters when material leaves the lab.
- Uses Google proposes: film pre-visualization, game-idea prototyping, and training scenarios for robots and AI agents, with enterprise partners testing fast scenario generation.
What it means for your business
- Prototype before building: the nearest business application is testing an idea in a walkable world before writing engine code, filming or building a set. A playable prototype shortens the conversation with a client or partner.
- Train without breaking anything real: for robots and agents, a simulated world lets you test scenarios that in the real world would cost time, money or risk. That is the training path DeepMind highlights.
- What to ask before buying: real useful session length, how fast the scene degrades, who owns the generated material, and how it is marked as AI-generated when published.
Want AI that works for your business?
Message us on WhatsApp and let's talk about automation and AI agents for your business.
WhatsApp inicIARelated articles
- Google launched the universal Gemini agent
- Google Agent Compute Engine: one microVM per agent
- How to use AI in your content without losing your voice
Frequently asked questions
What is Genie 4?
It is Google DeepMind's world model announced on October 27, 2026. It generates real-time interactive scenes, with native audio, that you can walk through and modify with text while they run.
Can I use Genie 4 already?
Not publicly yet. It is a research preview for invited testers, with access coming soon to Google AI Ultra subscribers through Project Genie and early-access partners. A waitlist is open.
How is it different from a video generator?
A video generator produces a clip to watch. A world model produces a place to walk through: you move, the camera responds, and you can change the scene while it runs.
Is Genie 4 the first world model with audio?
Google presents it that way for its line. Specialist coverage notes Runway released GWM Worlds 2 earlier, also playable and with audio, so the 'first' claim is best read with that nuance.
