In this article
Google DeepMind puts Genie 4 into Project Genie
Genie 4 is Google DeepMind's newest world model, and it is available in Project Genie. The new Project Genie experience is rolling out starting today to eligible Google AI Ultra subscribers.
Google says early testers wanted deeper control over what happens inside worlds and easier ways to build and share them. This update responds to that.
Genie 4 adds mid-session prompts and native audio
- You can type a new prompt mid-session to change the world in real time, for example to shift the weather, drop in a character or redirect the action.
- Genie 4 pairs its visuals with real-time native audio that adapts as the scene unfolds, from ambient soundscapes to character dialogue.
- Gemini models can join a session, watch the screen, offer suggestion chips for where to go next, or trigger in-world events automatically.
- A new Project Genie agent takes a natural-language description and orchestrates Google's multimodal models to create initial visuals and a matching custom background score.
- A community gallery offers curated worlds and community creations, and you can remix a world's structure, soundtrack and rules or publish your own.
Genie 3 stays in the model picker for longer runs or panorama-seeded spaces, while Google says Genie 4 gives the highest-fidelity visuals, physics and native audio.
Genie 3 gets longer sessions, chaining and panorama seeds
- Sessions can now run up to 3 minutes per generation, up from 60 seconds, and world chaining links multiple worlds together.
- 360-degree panorama image grounding lets you seed a new world from a panorama image.
- Real-time in-world events let you change the environment with natural-language prompts while exploring.
Genie 4 claims rest on Google's own post
Everything above comes from Google's announcement, where Google calls Genie 4 "our most realistic and interactive world model yet". Google also calls the real-time native audio a first. Google lists open challenges for Genie 3 and Genie 4: camera latency, intricate physical interactions, real-time audio precision and long-term scene permanence.
A separate Genie 4 research overview covers limitations. Treat the capability descriptions as vendor claims until someone outside Google tests them.
Genie 4 safety screening and SynthID watermarks
Google says all video and audio generated in Project Genie carries SynthID watermarking, and all model inputs and outputs go through safety screenings.
Project Genie runs in the browser for Ultra subscribers
Project Genie runs directly in the browser and is described as an experimental research prototype introduced earlier this year. Eligible Google AI Ultra subscribers get the rollout, and Google AI Pro members can join a waitlist on the same page. Before you plan around it, check Google's page for price, licence, API access, regions, usage limits and hardware needs.
Who should try Genie 4 and who should wait
If you already hold an Ultra subscription, this is a quick way to test prompt-driven worlds with audio in a browser tab. If you need something you can script, self-host or budget for, wait for Google to publish access details.
Sources
Image: Madhav-Malhotra-003, CC0 1.0, via Wikimedia Commons
This story was produced with AI assistance from the sources listed above and fact-checked against them before publication. Read our editorial methodology and the corrections log. Spotted an error? Email [email protected].

Discussion