Hot AI News

Google’s Project Genie now lets you type changes into a running world

Google is rolling out Genie 4, longer Genie 3 runs, a world-building agent and a community gallery in Project Genie to eligible AI Ultra subscribers; AI Pro users can join a waitlist.

Google’s Project Genie now lets you type changes into a running world
Image: Google / Google DeepMind

Google is rolling out a new Project Genie experience to eligible Google AI Ultra subscribers. It brings Genie 4, which lets users edit a world by typing new prompts while they move through it, plus a world-building agent and a community gallery.

Four additions inside one Project Genie update

According to Google’s announcement post, Genie 4 is now available in Project Genie. Google says users can change the environment in real time by typing a prompt during navigation. The model also generates native audio that adapts to the scene.

Still from an embedded demo video in Google's Genie 4 post. I did not check which demo it shows. The post's demos include a prompted spacewalk and a remixed cafe.
Image: Google / Google DeepMind

Genie 3 stays in the model picker and gets upgrades of its own. Runs can last up to 3 minutes per generation. Several worlds can be chained into one journey. A 360-degree panorama can seed a new world, and natural-language prompts can trigger events inside a world.

The Project Genie agent builds worlds from a conversation. Google says it coordinates multimodal models for the visuals and the background music. It can also bring Gemini into a session to watch the screen, suggest next steps or trigger events.

Still from a second embedded demo video in Google's Genie 4 post. I did not check which demo it shows.
Image: Google / Google DeepMind

The community gallery mixes curated and user-made worlds. Google names examples such as “Living Memory”, “Dino Field Guide” and “Rock Paper Scissors”, along with collaborations with outside creators. Users can remix gallery worlds and publish their own.

Advertisements

Who can try it, and what protections apply

Google AI Ultra subscribers can use the new experience at labs.google/projectgenie. Google AI Pro members can join a waitlist on the same page. The post gives no pricing, no list of countries and no usage caps for Genie 4. Google says generated video and audio carry SynthID watermarks and that all inputs and outputs go through safety screening.

The DeepMind Genie 4 post describes the model as the first real-time world model to generate audio natively, covering music, ambient sound and dialogue. It also says Gemini 3.8 Flash acts as a guiding agent inside the generation loop. That agent observes the scene, breaks goals into steps and adapts when the user intervenes. Google adds that Genie 4 is also coming to early-access partners in the Gemini API and Google AI Studio.

Gaps between the promise and what is documented

This is a research prototype, and Google’s own notes list real limits. DeepMind says Genie 4 is an early research model. Visual details and object permanence can drift in longer sessions or after several prompt edits. Complex physical interactions, real-time audio accuracy, legible text, spatial accuracy and long-term consistency are still open research areas. The post says only that it keeps environments coherent “over short sessions” and gives no duration, resolution or frame rate.

The comparison with the first release shows how much has moved. When Project Genie launched on January 29, 2026, Google’s post said it was open to AI Ultra subscribers in the US aged 18 and over. It was not available to AI Ultra for Business users, and generations were capped at 60 seconds. Promptable events that change a world mid-exploration were listed as not yet included. Those events now appear in both Genie 4 and Genie 3, and the Genie 3 cap has gone from 60 seconds to 3 minutes.

What the November post does not say is whether the US-only and 18-plus conditions still apply, or whether Business accounts are now covered. It uses the phrase “eligible” subscribers without defining it. Google’s claims about realism, audio quality and the agent’s ability to keep a plan on track come from its own posts, and no independent evaluation is cited.

For readers, the practical change is that Project Genie becomes less of a one-shot generator. Worlds can now be edited, extended, chained, scored and shared. Whether that is enough to draw people beyond the Ultra tier will depend on the Pro waitlist and on how Google handles access and limits for Genie 4.

Sign Up for the Daily AI Pulse

One email a day. All the stories that matter.

Alex CarterAI News & Big Tech Correspondent

Alex Carter writes for Aiholics, keeping readers updated on the fast-paced world of AI and Big Tech.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may also like