Google DeepMind's Genie 4 adds real-time 3D worlds with sound
Google DeepMind has released Genie 4, its most realistic world model so far, alongside longer Genie 3 sessions and a community gallery for sharing and remixing creations.

Key takeaways
- Genie 4 is Google DeepMind's most realistic world model so far, for Google AI Ultra subscribers.
- Genie 4 worlds can be edited in real time with text and include native sound.
- Genie 3 gains sessions of up to three minutes, 360-degree panoramas and chainable worlds.
- A community gallery allows sharing and remixing, with SynthID watermarks and safety checks.
In this article
What Project Genie now offers
Google DeepMind has updated Project Genie with a new model called Genie 4, according to The Decoder. The company describes it as its most realistic world model to date, and it is offered to Google AI Ultra subscribers.
Genie 4 produces interactive 3D worlds that can be changed while they run, using typed instructions. For the first time, the generated worlds also include sound.
Genie 3 gains longer sessions

The earlier Genie 3 model is receiving updates too, The Decoder reports. Sessions can now last up to three minutes, and the model supports 360-degree panorama views.
Genie 3 also gains chainable worlds, which let users link generated environments together. A new AI agent can build a world from a spoken description, and Google AI Pro members can join a waitlist.
Sharing, remixing and safety checks
A community gallery lets people publish the worlds they make and remix other users' creations. According to The Decoder, all generated content carries SynthID watermarks and passes through safety checks.
The Decoder does not give a release date for the wider rollout, and its report has not been independently confirmed by Google DeepMind.
Sources
Spotted a mistake? See our corrections policy. Read how News by OnDevice works in our editorial policy.



