Thursday, November 19, 2026

News

Genie 4 Arrives for Google AI Ultra Subscribers, With Worlds You Can Edit by Prompt Mid-Session

ModelsPatryk Raba
Genie 4 Arrives for Google AI Ultra Subscribers, With Worlds You Can Edit by Prompt Mid-Session
Fot. Gciriani, Wikimedia Commons (CC BY-SA 4.0)

Google is opening Genie 4 in Project Genie to eligible AI Ultra subscribers. The model adds native audio and lets users edit the world in real time.

Contents
  1. What Genie 4 changes
  2. Genie 3 stays in the model picker
  3. Agent and community gallery
  4. Limitations

Google DeepMind and Google Labs released Genie 4 on November 18 in a refreshed Project Genie. Eligible Google AI Ultra subscribers get access, and they can now type new prompts while exploring a generated world to change it on the fly.

What Genie 4 changes

According to Google's blog post, Genie 4 is the company's most realistic and interactive world model so far. During a session, a user can type a new prompt, for example one that changes the weather or adds a character, and the world changes in real time.

It is the first Genie model to generate sound together with visuals. The audio track, which covers ambient sounds and character dialogue, adapts to what is happening in the scene.

Genie 3 stays in the model picker

Genie 3 can still be chosen in the model picker. It received updates: sessions last up to 3 minutes per generation instead of 60 seconds, several worlds can be linked into longer journeys, and a world can be seeded from a 360-degree panorama.

Real-time in-world events triggered by natural-language prompts are also being added.

The new Project Genie agent builds worlds from a conversation: it creates initial visuals and background music matched to the mood. Gemini models can also be brought into a session, where they watch the screen, offer suggestions for where to go next as chips, and can trigger events on their own.

There is also a new gallery with worlds picked by Google and created by users. They can be browsed, remixed and published. Examples include "Living Memory," "Dino Field Guide" and "Villa Giallo," a mystery set in a gothic manor.

Limitations

Google calls Genie 3 and Genie 4 frontier research models. Among the unresolved challenges it lists camera latency, intricate physical interactions, real-time audio precision and long-term scene permanence. All inputs and outputs go through safety screenings.

The post does not list the countries where the feature is available or give any plans for an API. Users in Poland should therefore check availability on the Project Genie website. According to Google, early testers wanted more control over what happens inside the worlds, and mid-session prompt editing is the answer to that.

Share: