Google announced on November 18, 2026 (U.S. time), in an official blog post, that its world model Genie 4 is now available in Project Genie, its experimental service. Access is rolling out to eligible Google AI Ultra subscribers. Project Genie lets users describe a world in text and then move around inside the generated video as they explore it.
Genie 4 Reshapes Worlds Mid-Session and Generates Audio on the Fly
A world model is an AI model that keeps generating the next scene in real time in response to the user's actions. Google DeepMind describes Genie 4 as "our most realistic and interactive world model yet." Users navigate and control the camera with standard keyboard controls and steer the scene with text prompts.
With this update, Project Genie users can type a new prompt while exploring to change the world they created. Google gives examples such as shifting the weather, dropping in a character, or redirecting the action.
Audio is the other headline feature. Genie 4 generates sound alongside the visuals in real time, from ambient soundscapes to character dialogue. Google DeepMind says Genie 4 is its first real-time world model to generate audio natively.
Project Genie Agent and a Community Gallery Join the Service
Project Genie also adds the Project Genie agent, which lets users build a world through natural-language conversation. The agent orchestrates several of Google's multimodal models to create the initial visuals and a background score tailored to the mood. Google says this takes creation beyond a single text or image prompt.
During exploration, Gemini watches what is happening on screen, suggests where to go next, and can trigger in-world events automatically. Users can publish their worlds to a community gallery, where others can explore them or remix the underlying world structure, soundtrack, or rules. Gallery examples include Living Memory, which turns personal photos into explorable spaces, and Rock Paper Scissors, with Gemini acting as referee.
Genie 3 Now Supports Up to 3 Minutes per Generation
The earlier Genie 3 model remains available in the model picker. Google recommends it for longer runs or for spaces seeded from panorama images. The main differences are shown below.
| Item | Genie 4 | Genie 3 (after this update) |
|---|---|---|
| Changing the world with prompts while exploring | Supported | Supported |
| Audio | Real-time ambient sound and dialogue | Not stated |
| Exploration time per generation | Not stated | Up to 3 minutes (previously 60 seconds) |
| Other features | ― | Chaining multiple worlds / generating from 360-degree panorama images |
The longer runtime stands out. Exploration time per Genie 3 generation has grown from 60 seconds to up to 3 minutes. Google has not disclosed Genie 4's resolution, frame rate, or exploration time per session.
Ultra Subscribers First, With a Waitlist for Pro Members
As of this writing, Project Genie is rolling out to eligible Google AI Ultra subscribers. The Project Genie site lists the service as available to users 18 and older and describes it as an experimental research prototype. Google AI Pro members can join a waitlist.
Google did not say in this announcement which countries or regions are covered, whether the service is available in Japan, or when Pro members will get access. It also did not list a standalone price for Project Genie.
On safety, Google says all video and audio generated in Project Genie include SynthID watermarking. All model inputs and outputs also go through safety screenings.
Our Take
Google DeepMind released Genie 4 on October 27, 2026, as a research preview for a limited group of testers. Reaching paying subscribers about three weeks later suggests Google is moving quickly to turn this field into products. Generating audio together with visuals is a meaningful step for people who want to sketch out games or video projects.
That said, Google itself lists camera latency, intricate physical interactions, real-time audio precision, and long-term scene permanence as challenges. How long a single Genie 4 session can run is also unknown. Still, now that Ultra subscribers can try it hands-on, it should soon become clear how practical it really is.
Summary
Google has made Genie 4, a world model that lets users change worlds mid-session and generates audio simultaneously, available in Project Genie. It is rolling out to eligible Google AI Ultra subscribers, and Pro members can join a waitlist. Supported regions, including Japan, and the timing of Pro access had not been disclosed as of this writing.
Further reading: Google DeepMind's introduction to Genie 4