Google opens Genie 4’s interactive AI worlds to eligible Ultra subscribers
The Project Genie update adds a conversational creation agent and remixable gallery, while Google acknowledges unresolved limits in physics, audio precision and scene permanence.
Project Genie now lets eligible Google AI Ultra subscribers revise generated environments with text prompts while exploring, with Gemini helping create and guide experiences and audio that responds to the scene. The rollout also adds a gallery for sharing and remixing worlds, but Google still flags latency, complex interactions, precise audio, and scene persistence as research challenges. Genie 3 remains available for longer sessions and panorama-based worlds, while Google AI Pro subscribers can only join a waitlist, with no access date announced.
01
Genie 3 sessions can now run for up to three minutes per generation, and world chaining links multiple environments into longer journeys.
02
Genie 3 also supports worlds seeded with 360-degree panoramas and changed through natural-language events.
03
Google says Project Genie’s generated video and audio carry SynthID watermarks, and model inputs and outputs undergo safety screening.
Eligible Google AI Ultra subscribers can now start changing an AI-generated world while moving through it, rather than setting everything up before entering. Google began rolling out Genie 4 through Project Genie on November 18, 2026, pairing live text-directed changes with scene-responsive audio and adding tools to build and share those worlds.
Editing the scene from inside it
Project Genie is Google’s experimental browser-based tool for generating and exploring interactive environments. The updated experience combines its world models with Gemini models, letting users create a setting, enter it and shape what happens as they explore.
Google says Genie 4 accepts new text prompts during a session. Someone navigating a world can change the weather, introduce a character or redirect the action without leaving that exploration. Its native audio is generated in real time alongside the visuals, adapting to the unfolding scene with ambient sounds and character dialogue.
That responsiveness remains a research challenge. Google identifies camera latency, complex physical interactions, precise real-time audio and long-term scene permanence as areas still needing work. Those limits concern how the world responds and holds together, not just how convincing an individual view looks.
Google says all video and audio generated in Project Genie carry SynthID watermarks. Model inputs and outputs also undergo safety screening, safeguards that accompany both world models in the updated experience.
Genie 3 keeps a role for longer journeys
The new model does not replace every reason to use its predecessor. Google positions Genie 4 as its highest-fidelity option for visuals, physics and native audio, while keeping Genie 3 in the model picker for longer runs and worlds started from panoramic images.
Genie 3 also gains world chaining, which links multiple worlds into a longer journey. Users can seed an environment with a 360-degree panorama image and change it through natural-language events during exploration. Google demonstrates those features together in a flight across forests, underwater ruins and icy seas.
A conversation becomes a shareable world
The new Project Genie agent handles creation through conversation rather than a single text or image prompt. Google says it coordinates its multimodal models to produce the initial visual style and a custom background score matched to the requested mood.
Gemini can also participate after exploration begins. It observes what is on screen and offers context-aware suggestions for where to go next, or automatically triggers events inside the world. That gives the assistant a role during the experience as well as in its initial construction.
A new community gallery provides both ready-made places to visit and starting points for remixes. Users can work with the agent to alter an existing experience’s world structure, soundtrack or rules, then publish their creation for others to explore and remix. Google’s curated examples show several different approaches:
Living Memory turns personal photographs into explorable spaces.
Dino Field Guide uses panoramic images to ground prehistoric habitats.
Rock Paper Scissors sets themed objects against one another on a desk, with Gemini acting as a real-time referee.
Access starts with eligible Google AI Ultra subscribers. Google AI Pro members can join a waitlist through the Project Genie website, but the rollout announcement gives them no access date.
Reader comments
Newest comments first. Replies stay oldest first.