Friday, November 20, 2026

INFORMOPEDIA

World news and deep-dive explainers — climate, economy, technology, science, culture, health

Breaking
Technology

Google’s Genie 4 World Model Opens to AI Ultra Subscribers — Worlds That Talk Back

Google has opened its most realistic world model to the public for the first time. Genie 4, the system Google DeepMind first showed to trusted testers in late October, is now available to eligible Google AI Ultra subscribers inside a redesigned Project Genie experience, the company announced on Tuesday, November 18. Unlike a chatbot, which answers in text, Genie 4 builds a place you can walk through — and for the first time, you can change that place while you are inside it, and hear it respond.

Worlds you can rewrite as you explore them

The centrepiece of the update is mid-session editing. While navigating a generated world, a user can type a new prompt and the world changes around them — the weather shifts, a character appears, the action turns in a new direction — without the session restarting.

Genie 4 also pairs its visuals with real-time native audio for the first time. Ambient soundscapes and character dialogue are generated alongside the scene and adapt as it unfolds, so a world does not just look different when it changes — it sounds different too. Google calls Genie 4 its most realistic and interactive world model yet.

Genie 3 gets longer sessions too

The same update upgrades Genie 3, which remains available in Project Genie’s model picker. Sessions now run for up to three minutes per generation, up from 60 seconds, and multiple worlds can be chained together into longer journeys. Worlds can also be seeded from 360-degree panorama images, and Genie 3 gains real-time in-world events, so its environments can change on a natural-language prompt while you explore. Genie 3 stays the pick for longer runs and panorama-seeded spaces; Genie 4 is the pick for fidelity.

A conversational agent builds the world for you

Creation itself is changing. A new Project Genie agent takes over the first step: describe a vision in plain language and it orchestrates Google’s multimodal models to build the world’s opening visuals, matched with a custom background score set to the intended mood. Once inside, Gemini models can join the session — watching what happens on screen, offering suggestion chips for where to go next, or triggering in-world events on their own.

There is also a new community gallery, where curated worlds sit beside creations from the community. Visitors can turn personal photos into explorable spaces in “Living Memory”, wander prehistoric habitats in “Dino Field Guide”, or play “Rock Paper Scissors” with Gemini acting as referee. Two creator collaborations headline the launch: “Villa Giallo”, an atmospheric mystery inside a gothic manor built with creative technologist Will, also known as BLVCKL!GHT, and “The Life of Ants”, a journey through a colony from an ant’s perspective made with filmmaker Anand Gandhi. Worlds can be published to the gallery for others to explore and remix.

Who can try it — and what it costs

Access starts with eligible Google AI Ultra subscribers, who can try the new Project Genie experience from today. Google AI Pro members can join a waitlist on the same page. Google has not published a separate price for Project Genie access; it comes with a Google AI Ultra subscription.

The launch follows a busy November for Google’s AI division. The same week brought Gemini-powered smart glasses opening for pre-order and the Gemini 4 Neon model, suggesting the company is pushing its newest models into consumer products in parallel rather than one at a time.

Safety, and the limits Google admits

All video and audio generated in Project Genie carries SynthID, Google’s watermark for AI-generated media, and all inputs and outputs pass through safety screenings. But Google is clear that both Genie 3 and Genie 4 remain frontier research models, not finished products. It names four unresolved challenges: camera latency, intricate physical interactions, real-time audio precision, and long-term scene permanence — the tendency of a simulated world to drift or forget details over a long session. Independent reporting adds that visual details can drift during longer sessions, after complex edits, or while navigating, and that legible text and spatial accuracy remain active research areas.

Why it matters

World models are a different bet from the chatbot race. Instead of answering questions, they simulate places — and Google’s stated uses go beyond entertainment, including training environments for robots and AI agents that need to learn how to navigate the real world. Opening Genie 4 to paying subscribers turns a research preview into a product people can actually use, and the gallery turns it into a shared one. The test now is whether worlds that talk back stay coherent long enough for anyone to build something real inside them.

Avatar photo
Informopedia News Desk

Join the discussion

Your email address will not be published. Required fields are marked *