Google has released Genie 4, its newest interactive world model, inside Project Genie. The update is rolling out to eligible Google AI Ultra subscribers and adds live prompt editing, native audio, an agent that builds worlds from conversation, and a public gallery.
Project Genie is an experimental research prototype Google introduced earlier this year. It generates interactive worlds that people explore in the browser. Google says early testers asked for more control over what happens inside those worlds and easier ways to share them.
What Genie 4 can do
Google calls Genie 4 its most realistic and interactive world model so far. The headline feature is mid-session prompting. While you move through a world, you can type a new prompt to change it in real time: shift the weather, add a character, or redirect the action.
For the first time, Genie 4 also generates native audio that adapts as the scene changes, from ambient sound to character dialogue.
Genie 3 gets upgrades
Genie 3 stays in the model picker for longer runs and panorama-based worlds. Google lists three updates:
- Sessions can last up to 3 minutes per generation, up from 60 seconds, and worlds can be chained together for longer journeys.
- New worlds can start from 360-degree panorama images.
- You can trigger in-world events with natural-language prompts while exploring.
An agent and a gallery
The new Project Genie agent lets you describe a world in conversation instead of writing one prompt. According to Google, the agent coordinates its multimodal models to produce the opening visuals that set a world’s look, along with a custom background score matched to the mood.
Gemini models can join a session too. They watch what’s on screen, suggest where to go next through on-screen chips, or set off in-world events automatically.
The gallery collects curated and community worlds. Examples include “Living Memory,” which turns personal photos into explorable spaces, “Dino Field Guide,” which uses 360-degree panoramas of prehistoric habitats, and “Rock Paper Scissors,” where objects on a desk face off with Gemini acting as referee.
Google also worked with creators on early access projects. “Villa Giallo,” built with creative technologist Will aka BLVCKL!GHT, is a mystery inside a gothic manor. “The Life of Ants,” made with filmmaker Anand Gandhi, lets players move through a colony from an ant’s point of view.
Users can remix gallery worlds, changing the structure, soundtrack or rules, and publish their own for others.
Safety and known limits
All video and audio produced in Project Genie carries SynthID watermarking, and model inputs and outputs go through safety screening, Google says. The company still labels both models as frontier research and lists open problems: camera latency, intricate physical interactions, real-time audio precision and long-term scene permanence.
Why world models matter to Google
Google calls Genie 3 and Genie 4 world models, systems that generate an interactive environment that responds to a user’s actions and prompts rather than a fixed video. Project Genie pairs them with Gemini models so the same session can generate visuals, sound and game-like events.
The announcement frames the release as a response to tester feedback: more control inside worlds, and easier ways to build and share them. The agent and the gallery both address that second request. The mid-session prompt addresses the first.
The gallery’s remix feature also makes the product more social, because someone can start from another user’s world and adjust its rules or soundtrack.
Who can use it and what comes next
Access is limited to eligible Google AI Ultra subscribers for now. Google hasn’t announced a wider date or pricing change. For people who build games, film previews or interactive demos, the practical test is whether a world holds together past a few minutes, since Genie 3 sessions still cap at 3 minutes.
Watch for Google’s progress on the limits it named, especially scene permanence, which decides whether these worlds can grow beyond short experiments.
Source: Google