subagentic.ai RSS X YouTube GitHub
AI-generated illustration

News

Genie 4 opens in Project Genie with a world-building agent

Google opens Genie 4 in Project Genie for AI Ultra, with live audio, mid-session edits, and an agent that builds worlds from conversation.

google-deepmindgenieproject-genieworld-models

Google is rolling out a new Project Genie experience to eligible Google AI Ultra subscribers, with first-time access to Genie 4. The November 18, 2026 post also adds a conversational agent that builds a world from a description, and a community gallery. Google AI Pro members can join a waitlist on the same page.

Project Genie is the experimental prototype Google introduced earlier this year for generating and exploring interactive worlds from the browser. The company says early testers wanted deeper control inside those worlds and easier ways to build and share them. This update puts Google’s world models alongside its Gemini models.

Genie 4: edit the world while you are in it

Google calls Genie 4 its most realistic and interactive world model yet. It is also the model the post credits with the highest-fidelity visuals, physics, and native audio. Type a new prompt while you navigate and the environment changes in real time: shift the weather, drop in a character, or redirect the action. For the first time, Genie 4 pairs those visuals with real-time native audio that adapts as the scene unfolds, from ambient soundscapes to character dialogue.

The post does not give Genie 4 a session-length cap.

The agent sets the look and the score

The Project Genie agent takes creation past a single text or image prompt. Describe a vision in natural language, and the agent orchestrates Google’s suite of multimodal models to craft the initial visuals — the world’s signature look — and a custom background score matched to the mood.

@Google posted the same description that day: building an interactive world is as easy as a conversation, the agent sets the signature look and custom soundtrack, and Gemini watches what is on screen to surface context-aware suggestion chips.

The blog adds that you can bring Gemini models into the session as you explore. It observes the screen, surfaces those chips for where to go next, or triggers in-world events automatically. You can also remix a gallery world with the agent — structure, soundtrack, or rules — and publish the result for others to explore and remix.

Genie 3 stays in the model picker

Genie 3 remains available for longer runs or panorama-seeded spaces. Updates listed in the post:

  • Up to 3 minutes per generation, up from 60 seconds, or world chaining that links multiple worlds
  • 360-degree panorama image grounding, so a new world can be seeded from a panorama
  • Real-time in-world events from natural-language prompts while you explore

What you can publish, and what is still open

The gallery mixes curated worlds with community work, and it is where creations can be explored, remixed, and published. Named examples include “Living Memory” (personal photos as explorable spaces), “Dino Field Guide” (prehistoric habitats from 360-degree panoramas), and “Rock Paper Scissors,” with Gemini as a real-time referee. Early collaborations called out are “Villa Giallo,” a gothic-manor mystery with creative technologist Will (BLVCKL!GHT), and “The Life of Ants,” with filmmaker Anand Gandhi, using Genie 4’s photorealism to move through a colony from an ant’s perspective.

All generated video and audio include SynthID watermarking. Inputs and outputs go through safety screenings.

Google says Genie 3 and Genie 4 remain frontier research world models. Open problems it names: camera latency, intricate physical interactions, real-time audio precision, and long-term scene permanence.

Eligible Ultra subscribers can try the new Project Genie experience; Pro members can join the waitlist, per the November 18 post. Start with that Keyword article for the full claims, then the @Google note for the company’s short version of the agent and on-screen suggestions.

Sources