Summary:

Google began rolling out Genie 4 to eligible Ultra subscribers on November 18. Genie 4 adds native audio, while Genie 3 still offers longer runs and panorama starts.

Google began rolling out Genie 4 to eligible Google AI Ultra subscribers on November 18. The model arrived inside Project Genie, Google’s browser-based world-building prototype, after an October research preview for trusted testers. You can steer an explorable scene with text while moving through it, and Genie 4 can generate sound along with the world. For longer runs, Google’s updated Genie 3 is still in the model picker.

Who can try Genie 4?

Open Project Genie in a browser and sign in with your personal Google Account. Google says the rollout starts with eligible Google AI Ultra subscribers in places where the plan is offered, with an age minimum of 18. Google AI Pro subscribers can join a waitlist; a waitlist entry is not immediate access. Usage limits can vary with plan and demand. The November 18 announcement covers the Project Genie experience, not general API availability.

When to pick Genie 4 over Genie 3

CapabilityGenie 4Updated Genie 3
Exploration durationAbout one minute of continuous interactionUp to three minutes per generation; worlds can be linked
Events while exploringSteer the environment with live text promptsTrigger in-world events with natural-language prompts
What stands outNative scene audio, including ambient sounds and character dialogue360-degree panorama seeding and linked worlds for longer routes

Try Genie 4 when a short scene needs to respond and sound convincing, such as testing how a quiet walk changes when rain starts. Choose Genie 3 for a longer route or when you want to start from a 360-degree panorama. Google’s model card reports qualitative improvements in realism, motion and steering over Genie 3; those findings are distinct from the product’s session limits.

What the project agent and gallery add

The Project Genie agent can help turn a description into a starting world and soundtrack. The gallery brings together curated and community worlds that users can explore or remix. Those are features of the Project Genie experience, separate from Genie 4’s ability to generate sound as the scene changes.

DeepMind notes that objects and details can drift, especially during longer sessions or after several complex edits. Real-world locations, legible text, complicated physics and precise audio remain imperfect. Google says video and audio generated in Project Genie carry its SynthID watermark.

Sources