Money and business in the Middle East.

AI

Google opens Genie 4 world model to paying AI Ultra subscribers

Google released Genie 4 inside Project Genie to eligible AI Ultra subscribers starting November 18

· Last verified: 19 Nov 2026 (Google DeepMind)

Summary

  • Google released Genie 4 inside Project Genie to eligible AI Ultra subscribers starting November 18
  • The model generates interactive worlds with native real-time audio, editable by typed commands mid-session
  • Access moves from a closed trusted-tester group to a paid consumer tier

The latest

Genie 4, Google's most realistic world model by its own description, became available on November 18, 2026 to eligible Google AI Ultra subscribers through the Project Genie experience. Users can type instructions during a live session to change weather, add a character or redirect events. Google AI Pro members can join a waitlist on the same page.

Details

  • The authors: The announcement post, titled Genie 4 is now available, was signed by Diego Rivas, group product manager at Google DeepMind, and Pavan Kumar, group product manager at Google Labs. It contained no direct quotes from either signatory.
  • What ships: The experience bundles four pieces: the Genie 4 model, an updated Genie 3, a Project Genie agent, and a gallery of curated and community-made worlds. Project Genie was introduced earlier in 2026 as an experimental research prototype rather than a product.
  • The audio first: According to Google, Genie 4 is the first Genie model to pair generated visuals with native real-time audio that adapts to the scene, covering ambient environmental sound and character dialogue. Google calls it its most realistic and interactive world model to date.
  • Genie 3 updates: Genie 3 stays selectable in the model menu and gains longer sessions: exploration now runs up to 3 minutes per generation instead of 60 seconds. Users can chain worlds into longer journeys, build from 360-degree panoramic images, and alter environments with natural-language commands while exploring.
  • The agent: The Project Genie agent builds worlds through natural-language conversation, coordinating several Google multimodal models to produce the opening image and mood-matched background music. During exploration, Gemini can watch the screen, suggest destinations and trigger in-world events automatically.
  • The gallery: Users can publish their own worlds and remix existing ones by changing structure, music or rules. Google cited Living Memory, which turns personal photos into explorable spaces; Dino Field Guide, prehistoric environments built from 360-degree panoramas; and Rock Paper Scissors, a desktop game with Gemini as referee.
  • Creator collaborations: Two early-access projects were named. Villa Giallo, made with creative technologist Will, known as BLVCKL!GHT, is a room-to-room puzzle set in a gothic mansion built with Genie 4. The Life of Ants, with director Anand Gandhi, is a walkthrough of an ant colony relying on Genie 4's photorealism.
  • Safety markers: All video and audio generated in Project Genie carries a SynthID watermark, and Google said every input and output passes through safety checks. The post did not list which countries can access the experience, nor set a separate price beyond the subscription tiers.
  • Stated limits: Google said Genie 3 and Genie 4 remain frontier research models, listing four areas still under work: camera response latency, complex physical interactions, real-time audio accuracy, and long-horizon scene consistency.

Background

Google announced Genie 4 on October 27, 2026 as a research release limited to a small group of trusted testers, promising wider availability soon to AI Ultra subscribers through Project Genie and to early-access partners via the Gemini API and AI Studio.

Between the lines

The three-week gap between the trusted-tester announcement and paid availability shows Google moving a stated frontier research model into a consumer subscription while its own list of unresolved issues — latency, physics, audio accuracy, scene stability — remains open. The SynthID watermark on every generated clip signals that distribution of synthetic worlds, not just their quality, is a governing concern.

What's next

Watch whether AI Pro waitlisted users get access, whether Google publishes country availability, and whether early-access partners ship Genie 4 projects through the Gemini API and AI Studio.

Source: Google DeepMind