Google rolls out Genie 4 world model, Project Genie agent and community gallery to Google AI Ultra subscribers
Edited by Elliot Marsh, Editor · written with AI from the linked sources (how)
Google began rolling out a new Project Genie experience to eligible Google AI Ultra subscribers. It includes the Genie 4 world model (real-time prompt-based editing of worlds mid-session, plus native adaptive audio), updates to Genie 3 (sessions up to 3 minutes, up from 60 seconds; world chaining; 360-degree panorama grounding; real-time in-world events), a Project Genie agent that builds worlds through conversation and brings Gemini into the session, and a community gallery for publishing and remixing worlds. Google AI Pro members can join a waitlist. Outputs carry SynthID watermarking.
Entities: Google, Google DeepMind, Google Labs, Project Genie, Genie 4, Genie 3
1 primary
What happened
Google has started rolling out a new Project Genie experience to eligible Google AI Ultra subscribers, announced on its own Gemini blog. It adds the Genie 4 world model, which lets users edit a generated world in real time mid-session and adds native audio that adapts to the scene. Genie 3 also gets updates: sessions run up to 3 minutes (previously 60 seconds), and it gains world chaining, 360-degree panorama grounding and real-time in-world events. A Project Genie agent builds worlds through conversation and brings Gemini into the session, and a community gallery lets users publish and remix worlds. Google AI Pro members can join a waitlist, and outputs carry SynthID watermarks.
Why it matters
For now this is a consumer and enthusiast product, limited to paying Ultra subscribers, so the direct audience is small. Researchers and competitors building interactive world models get a public benchmark for what Google will ship: longer sessions, editing during a session and audio. A three-minute limit still means these are short demos, not usable game or simulation environments. Developers should note that nothing here mentions an API or developer access, so it is unclear whether this becomes something they can build on.
What is noise
"Most realistic and interactive world model yet" is Google's own claim, with no independent benchmark or side-by-side comparison offered. Words like "responsive, creative and social" are packaging; the concrete gains are the 3-minute cap and mid-session editing. Missing context includes pricing, usage limits, compute cost, how well worlds stay consistent over three minutes, and any third-party testing.
Watch next
- 01Independent hands-on tests and videos from Ultra users showing whether worlds stay coherent across the full 3 minutes, after edits, and when chained.
- 02Whether Google widens access beyond Ultra (the Pro waitlist converting to real access, more countries) or opens an API or developer tools for Genie.
- 03How the community gallery is used and moderated in the first weeks, including how well SynthID watermarking holds up, and whether rivals respond with comparable interactive world models.
Coverage
2 storiesNewsletter
Get the no-hype AI news straight to your inbox
Weekly. No spam, unsubscribe any time. Privacy
More capability signals
Full feed →- Deepseek releases V4.1-Flash, an open-source model that sharply cuts KV cache memory and input-processing compute for AI agents10 Sept 202682
- Mistral releases Mistral Large 4, a ~1T-parameter open-weight multimodal MoE model, with day-0 Hugging Face support29 Oct 202680
- Meta releases Muse Spark Open, an open-weight ~816B-parameter multimodal mixture-of-experts model, with Hugging Face support15 Oct 202680
- OpenAI discloses sandbox-escape and credential-leak incidents, confirms pause on tool-use for its most capable models26 Sept 202680