Free Intensive Nov 21–22Free 2-day hands-on Vibecoding Intensive on Nov 21–22 at 16:00 MSK (Sat–Sun) · Live Cohort 3 starts Nov 23Join Free →Cohort 3 (Nov 23)
ReleaseAPI / Product · ActionableThe Decoder→ original

Google DeepMind Launches Genie 4: Text Reshapes 3D Worlds With Sound On the Fly

Google DeepMind has released Genie 4, a model that builds a walkable 3D world from a description. Mid-session, a user can type "let it start raining," and the scene rebuilds on the fly while the audio adjusts to match the picture. The model was previously a research preview, and subscribers to Google AI Ultra now have access. For game designers, it offers a way to sketch scenes quickly with words.

AI-processed from The Decoder; edited by Hamidun News
Google DeepMind Launches Genie 4: Text Reshapes 3D Worlds With Sound On the Fly
Source: The Decoder. Collage: Hamidun News.
Listen to article
Article Fact Sheet & Key Figures
ParameterDetails
Primary sourceThe Decoder ↗
Editor-in-Chief
Key metric #1Genie 3 has been updated with sessions of up to three minutes and 360-degree panoramas
Key metric #2Worlds can be chained together, and an AI agent assembles them from voice input
Key metric #3A gallery supports remixing, and content carries SynthID watermarks

Imagine saying "a rainy street in a city at night" and, within seconds, walking down it like a level in a video game. Then you type "let the sun come out," and the weather changes while you're still strolling. This is the scenario Google DeepMind is building toward with Project Genie. On November 18, the company released a major update centered on Genie 4.

What Genie 4 can do

DeepMind calls Genie 4 its most advanced "world model" for generating interactive 3D worlds. A world model is a neural network that maintains an entire scene in memory: the user moves through it, and the scene stays coherent. A video generator produces a finished clip, but here you are inside the action. Until now, Genie 4 existed as a research preview. It is now open to subscribers of the Google AI Ultra plan.

The main feature is real-time control of the world through text. Within a session, you can change the weather, the characters, or the storyline. There's no need to restart generation from scratch: the edits apply to the existing scene. The second new feature is built-in sound, which is also generated on the fly. When rain starts or a new character appears, the audio changes along with the image.

What Genie 3 gets

The previous version has also been updated. In Genie 3, sessions can now run for up to three minutes. Support for 360-degree panoramas has been added too, so a world can be viewed in full or shown to colleagues as a tour. There are also "world chains" now. One generated world links to the next, so instead of a single isolated scene, you get a longer journey.

For anyone who doesn't want to describe everything by hand, DeepMind has added an AI agent. You tell it by voice what you want to see, and the agent assembles a starting scene from that description. A shared gallery has launched alongside it. Users publish their work and make remixes, taking someone else's starting point and reworking it to their own taste.

Safety and access

All generated content carries SynthID watermarks and goes through safety checks. For interactive worlds that are easy to share and easy to modify, this matters. Remixing quickly blurs authorship, and the watermark helps distinguish generated content from captured footage.

Access is tightly controlled. Genie 4 is open to Google AI Ultra subscribers, the top paid tier. Google AI Pro owners can join a waitlist. The announcement includes no free access.

What this means in practice

For game developers, product designers, and prototype authors, Genie is for now more of a sketchbook than a replacement for a game engine. Three minutes of a Genie 3 session yield a sketch, not a finished level. It is too early to build a production pipeline on it. But for quickly visualizing ideas, the model looks useful. In minutes, you can check a scene's atmosphere, hear how it sounds, change the weather or the plot, and show the result to your team.

There is also a takeaway for the industry. World models are becoming a product with subscriptions, a gallery, and a community, and they are no longer just demos attached to a research paper. Now that Genie supports remixing, DeepMind is counting on user-generated content, as social networks and game modding communities have done. Those working with robots and simulations should also take note. Interactive worlds that change on command resemble training simulators, although the announcement says nothing about this.

If you have an Ultra subscription, it's worth spending an evening testing your own ideas to see where the model holds a scene steady and where it falls apart. Everyone else has the waitlist and can watch how quickly DeepMind expands access.

Story

Key Questions Answered

What Genie 4 can do?

DeepMind calls Genie 4 its most advanced "world model" for generating interactive 3D worlds. A world model is a neural network that maintains an entire scene in memory: the user moves through it, and the scene stays coherent. A video generator produces a finished clip, but here you are inside the action. Until now, Genie 4 existed as a research preview. It is now open to subscribers of the Google AI Ultra plan. The main feature is real-time control of the world through text.

What Genie 3 gets?

The previous version has also been updated. In Genie 3, sessions can now run for up to three minutes. Support for 360-degree panoramas has been added too, so a world can be viewed in full or shown to colleagues as a tour. There are also "world chains" now. One generated world links to the next, so instead of a single isolated scene, you get a longer journey. For anyone who doesn't want to describe everything by hand, DeepMind has added an AI agent.

What are the key details on Safety and access?

All generated content carries SynthID watermarks and goes through safety checks. For interactive worlds that are easy to share and easy to modify, this matters. Remixing quickly blurs authorship, and the watermark helps distinguish generated content from captured footage. Access is tightly controlled. Genie 4 is open to Google AI Ultra subscribers, the top paid tier. Google AI Pro owners can join a waitlist. The announcement includes no free access.

ZK
Zhemal Khamidun — CPO at Alpina Digital, product manager of AlpinaGPT. Building AI products since 2019. Hamidun News is a pet project that grew into a production aggregator with 6,000+ readers.
✦ Private AI Builders Community · Vibe Club

Ship AI Faster with Founders, Engineers & Weekly Architecture Reviews

Inside Vibe Club (hamidun.com/club) — weekly live masterminds with Zhemal Khamidun, Tech Radar, full LMS access, and 370+ production Claude Code skills.

Join Vibe Club →Explore Hamidun Academy
What do you think?
Loading comments…