Google DeepMind opens its Genie 4 world model to AI Ultra subscribers
Google is rolling out Genie 4 in Project Genie for eligible Google AI Ultra subscribers, with live prompt edits, real-time audio, a creation agent and a community gallery.
Google DeepMind and Google Labs began rolling out a new version of Project Genie on November 18, 2026, giving users their first access to the Genie 4 world model. Project Genie generates interactive worlds that you can move through in a browser. According to The Decoder, Genie 4 had only been available as a research preview before.
What's new
- Edit while you explore: Google says you can type a new prompt during a session to change the world, for example the weather, a new character or the direction of the action.
- Native audio: Genie 4 adds real-time sound that follows the scene, from ambient noise to character dialogue, according to the announcement.
- Genie 3 updates: the older model stays in the model picker. Each generation can now last 3 minutes instead of 60 seconds, plus chained worlds and worlds seeded from 360-degree panorama images.
- Project Genie agent: a new agent builds a world from a conversation, using Google's multimodal models for the first visuals and a background score.
- Community gallery: users can publish their worlds for others to explore and remix.
Availability and pricing
Google says eligible Google AI Ultra subscribers can try it now, and Google AI Pro members can join a waitlist. All video and audio generated in Project Genie carries SynthID watermarks, and model inputs and outputs go through safety screening.
Why it matters
Analysis: world models have mostly been research demos. Putting Genie 4 into a paid plan, with sharing and remixing built in, tests whether people will treat generated worlds as something to create and pass around, the way they already do with AI images and video. Google itself lists open problems, including camera latency and keeping scenes consistent over long sessions, so this is still an early product.
Source: Google