Newsroom
19 November, 2026 / News / AI / Tags: genie, google, worlds, audio, ultra

Google DeepMind has opened access to its latest world model through Project Genie, enabling eligible AI Ultra subscribers to edit generated environments while exploring them with real-time audio and conversational characters
Google DeepMind has begun making Genie 4 available to eligible Google AI Ultra subscribers via Project Genie, an experimental browser-based platform for generating and interacting with virtual worlds. The update builds on the model’s October unveiling, allowing users to change scenes mid-exploration by entering text prompts while navigating the environment.
Genie 4 generates interactive worlds on the fly rather than rendering finished videos. Users can shift the weather, introduce new elements, or redirect action without restarting the session. The model pairs these visual changes with native audio that includes ambient sounds and dialogue, which adapts to events or user input within the scene.
Users can now add characters to the generated scenes that respond to their presence and actions. These talking characters bring added depth to the exploration experience, as dialogue reacts to what happens in the environment.
The rollout also incorporates a Project Genie agent that assists users in creating initial worlds. By describing a vision in natural language, the agent helps establish visuals and custom background scores tailored to the desired mood.
Project Genie now includes a community gallery where users can discover curated worlds, explore others’ creations, and remix existing experiences. The platform features an AI creation assistant and tools for sharing and discovering user-generated content.
Genie 3, the previous-generation model, remains available alongside Genie 4. The earlier model has received updates that extend sessions to three minutes and support 360-degree panorama image grounding for more consistent spaces.
| Model | Key Features | Session Length |
|---|---|---|
| Genie 4 | Real-time editing, native audio, conversational characters | Not specified |
| Genie 3 | Longer runs, panorama grounding, real-time promptable events | Up to 3 minutes |
Project Genie originally launched in January 2025 as an experimental tool for adult Ultra subscribers in the United States. Early versions offered short 60-second sessions with limited reliability in character controls and physical behavior. The October 2026 introduction of Genie 4 marked a significant leap in realism, interactivity, and audio capabilities.
Google plans to make the update available to eligible Google AI Ultra subscribers. The earlier research preview was open to trusted testers, and subscriber access had been planned through Project Genie. Google DeepMind unveiled Genie 4 on Oct. 27 as a model for generating interactive worlds with audio.
Google acknowledges that the technology remains in the frontier research space. Challenges such as camera latency, intricate physical interactions, real-time audio precision, and long-term scene permanence continue to be addressed through ongoing research. Objects in generated scenes can still drift between views, and complex physics or generated speech may not always be accurate.
The platform uses keyboard and mouse controls for movement, and the model predicts what the world should look and sound like as changes occur. This approach differs from traditional game engines that rely on fully defined rules and stable levels. Google has not announced when access will extend beyond AI Ultra subscribers to the AI Pro tier, which is now on a waitlist.
Google notes that OpenAI and Meta AI are also working on similar world models, reflecting growing industry interest in these technologies.




