AI Models

Google opens Genie 4 world model to subscribers

BUZZINGA team5 min read

On November 18, Google began rolling out the Genie 4 interactive world model to eligible Google AI Ultra subscribers via the experimental browser-based Project Genie platform. The update adds real-time scene editing, synchronized native audio, an intelligent world-building agent, and a community gallery for sharing and remixing.

Google has moved its newest virtual world-building model, Genie 4, from closed testing to Ultra subscribers via the browser, delivering a qualitative leap that allows users to adjust weather, characters, and plot trajectories in real time while navigating inside the scene, featuring interactive native audio and an AI agent that helps direct events.

On Wednesday, November 18, Google began rolling out Genie 4, its latest interactive world-building model, to eligible Google AI Ultra subscribers through Project Genie, its experimental web browser-based prototype. AI Pro tier members can sign up for the waitlist through the same page, according to Google’s official announcement and a report by Seeking Alpha. This release moves the Genie 4 model beyond the small group of “trusted testers” who gained access late last month, as reported by The Verge, noting that Google first previewed the model to testers in October. This update adds real-time editing features, synchronized audio, a dedicated world-building tool, and an integrated way to share creations with others.

Real-time virtual world editing during sessions and Genie 3 updates

Google describes the Genie 4 release as its “most realistic and interactive world model yet.” Users can type new text commands while moving inside the generated world to change weather conditions, add a new character, or redirect the course of events without pausing the experience. This release is also the first model in the Genie series to support native audio generated in real time, ranging from ambient background sound effects to character dialogue, with the soundscape automatically shifting as visual scene details change.

Google explained that it has also updated the older Genie 3 model, which remains available within the platform’s model selection menu; sessions in it can now last up to 3 minutes per generation instead of the previous 60 seconds, and users can now link multiple sequential worlds together and start new virtual worlds from 360-degree panoramic images.

An AI agent for building worlds and directing live events

The new Project Genie agent allows users to describe the world they envision using simple natural language, after which it relies on Google’s multimodal models to create initial visual scenes and a custom background musical track suited to the environment’s mood. The company noted that the Gemini model can also join the live session, observing the screen, suggesting next steps, or triggering surprise events within the virtual world. Google gave an example of the agent adding an avalanche in the middle of a ski track the user is navigating.

Interactive community gallery and shared artistic experiences

A new community gallery blends curated worlds with those designed by users themselves, granting them the ability to publish their own creations and remix and edit the work of others. Prominent projects on display include “Villa Giallo,” a mystery puzzle set inside a Gothic mansion designed in collaboration with tech creator Will, known as BLVCKL!GHT, alongside the “The Life of Ants” project completed in collaboration with filmmaker Anand Gandhi, displaying an entire ant colony from the ant’s own perspective.

Current research limitations, security controls, and watermarking

Alphabet’s Google emphasizes that both Genie models remain within cutting-edge research systems under development, as its teams continue working on addressing camera latency, complex physical interactions, audio accuracy, and maintaining scene stability and consistency across long sessions. The company added that all video and audio generated within Project Genie carry Google’s invisible SynthID watermark, and all inputs and outputs undergo safety and security checks. The Next Web noted that visual details are still prone to drifting or changing during long sessions, and readable text and complex physics remain difficult challenges for the model. Google DeepMind concluded its statement by saying this release “brings native audio, real-time editing, and a new agent to help you build interactive virtual worlds.”

Frequently asked questions

Question: Who can currently use the Genie 4 model via the Project Genie platform?
Answer: The model is currently available to eligible Google AI Ultra subscribers via web browser, while AI Pro subscribers can sign up for the official waitlist to gain access later.

Question: What are the most prominent new interactive capabilities introduced by Genie 4 compared to previous versions?
Answer: Genie 4 allows users to type text commands while exploring the world to change weather, add characters, or modify plot events instantly, and it is the first release to generate native audio and dialogue that react in real time to scene changes.

Question: What improvements did the older Genie 3 model receive in this update?
Answer: Google increased the duration of a single session in Genie 3 from 60 seconds to 3 minutes, with support for linking sequential worlds together and creating new environments from 360-degree panoramic images.

Question: What technical limitations still face the Genie models and how does Google regulate content safety?
Answer: The models still face challenges with camera lag, complex physics, readable text, and long-term detail consistency. Google integrates the SynthID watermark into every audio and video file alongside comprehensive screening of all inputs and outputs.