Google DeepMind brings Genie 4 to Project Genie for interactive worlds

Google DeepMind is rolling out Genie 4 through an updated Project Genie experience for eligible Google AI Ultra subscribers. The release adds real-time world editing, adaptive native audio, a creative agent and a gallery where users can explore, publish and remix interactive worlds.
- Genie 4 enables real-time text-based edits to generated worlds.
- Native audio adapts to scenes, including ambience and character dialogue.
- Genie 3 now supports runs of up to 3 minutes and linked worlds.
- A Project Genie agent helps users build worlds through conversation.
- Eligible Google AI Ultra subscribers can try the experience today.
Genie 4 adds live editing and native audio
Project Genie is an experimental research prototype for generating and exploring interactive worlds in a browser. With Genie 4, users can type prompts while navigating a world and change it during the session. Google DeepMind says they can shift the weather, add a character or redirect the action without leaving the experience.
The model also pairs generated visuals with real-time native audio. Audio can adapt as a scene develops, including ambient soundscapes and character dialogue. Google describes Genie 4 as its most realistic and interactive world model yet, with its highest-fidelity visuals, physics and native audio.
The updated experience is intended to give users more control over worlds after they have been generated. It combines Google’s world models with Gemini models, allowing people to generate, enter and shape environments directly from a browser.
Genie 3 supports longer journeys and panorama-based worlds
Genie 3 remains available in Project Genie’s model picker for users who want longer runs or spaces seeded from panorama images. Google says each generation can now run for up to 3 minutes, compared with 60 seconds previously. Users can also link multiple worlds together to create longer journeys.
Other Genie 3 updates include 360-degree panorama image grounding and real-time in-world events. Natural-language prompts can dynamically change an environment as users explore, allowing connected journeys through settings such as forests, underwater ruins and icy seas.
Google says panorama grounding can help keep elements such as Petra’s canyon and facade consistent as users look around. The company also presents examples involving longer flights through generated environments and events summoned on command.
An agent and gallery broaden creation and sharing
The new Project Genie agent lets users describe an idea in natural language instead of relying on a single text or image prompt. According to Google DeepMind, the agent coordinates Google’s multimodal models to create initial visuals, establish a world’s look and produce background scores suited to its mood.
Gemini models can also observe what is happening on screen, offer context-aware suggestions for where to go next and trigger in-world events automatically. Users can publish their worlds to a new community gallery, or remix existing creations by changing their structure, soundtrack or rules.
The gallery includes curated and community-built experiences such as “Living Memory,” “Dino Field Guide” and “Rock Paper Scissors.” It also features “Villa Giallo,” created with Will aka BLVCKL!GHT, and “The Life of Ants,” built with filmmaker Anand Gandhi.
Eligible Google AI Ultra subscribers can try the updated Project Genie experience starting today. Google AI Pro members can join the waitlist. Google says all video and audio generated in Project Genie includes SynthID watermarking, while model inputs and outputs go through safety screenings. The company also says Genie 3 and Genie 4 remain frontier research models, with ongoing work on camera latency, complex physical interactions, real-time audio precision and long-term scene permanence.
Google DeepMind: Genie 4 is now available

