Google DeepMind opened a research preview of Genie 4 to trusted testers on Tuesday, adding sound and responsive speech to its interactive AI worlds. Experiences last about two minutes at this week's launch, and users can change a running scene through free-form text instructions. Google says filmmakers and other trusted testers are trying it to draft scenes and test different camera angles or dialogue before final production.

The two-minute experiences compare with the 60-second limit in the earlier, Genie 3-powered Project Genie consumer app introduced in January. A world model generates the next sights and sounds as someone moves through an environment.

What Changed

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Drafting a scene

Google DeepMind engineer Diego Rivas described the appeal for filmmakers as “keeping a state of flow” during early creative work and storyboarding. In Google's account, testers can enter a rough simulation and run successive takes, adjusting lighting or changing what characters say while the scene continues. A filmmaker can try a different camera angle within the running draft, then change the dialogue without restarting the scene.

Google describes this as an experimental draft stage. Rivas said turning those prototypes into creative assets with Gemini Omni was a possible future workflow, with those capabilities still some way off. The company has not settled what the complete product experience will be.

Gemini checks the scene

Users move through the generated environment with keyboard controls. Text instructions can alter the weather or camera position, or introduce people and objects without restarting the scene. Google says the model generates ambient sound and music along with dialogue, with characters responding to user inputs and events. The audio is generated alongside the imagery, so sound is part of the simulated scene as it runs.

In Google's explanation, Gemini Omni helps scenes look and behave like the real world. Gemini 3.8 Flash watches the generated images, divides a goal into steps and checks new images to see whether each step succeeded. When a user intervenes or the scene changes, Flash adjusts its plan to account for what is now happening in the scene.

Google also claims more realistic human movement and improved handling of physical interactions such as collisions.

Details can drift

Google says visual details and object permanence can deteriorate as sessions grow longer, particularly after repeated complex edits or movement through a scene. Object permanence means an object should remain there as someone moves through the environment. In the preview, Google warns, objects may change or fail to stay consistent over time, even as the surrounding scene continues. Accurate physics and real-time audio remain research problems, as do readable text, spatial accuracy and continuity over longer periods.

Know someone who'd find this useful? ✉️ Email it to a friend in one click, or they can subscribe free here.

Reporter Jay Peters had not tried the new model when he covered its announcement on October 27.

His January 29 hands-on with the earlier, Genie 3-powered Project Genie app exposed visible continuity failures. Paint trails disappeared after a ball rolled over snow, and part of a racetrack unexpectedly became grass. Those were results from the old consumer app, not a test of Genie 4.

Wider access is pending

Google says every generated video and audio track will carry SynthID, its imperceptible watermark for identifying AI-generated material.

A waitlist is open for Genie 4 in Project Genie. Google plans to expand access through that app to subscribers of AI Ultra, its subscription tier, and to early partners through the Gemini API and Google AI Studio. Its announcement describes those releases as coming soon, without a firm date.

Frequently Asked Questions

What does Genie 4 generate?

It generates interactive worlds with imagery, ambient sound, music and responsive dialogue. Users can move through them and change a running scene with text instructions.

How long do the experiences last?

About two minutes at the October 27 launch, compared with the 60-second sessions in the earlier, Genie 3-powered Project Genie consumer app.

How is Gemini involved?

Google describes Gemini Omni as grounding appearance and behavior. Gemini 3.8 Flash watches frames, divides goals into steps, checks progress and adjusts plans when users intervene.

What limitations remain?

Google says visual details and object permanence can drift during longer sessions or complex edits. Accurate physics, real-time audio, readable text, spatial accuracy and longer-term continuity remain research problems.

Who can use Genie 4?

The research preview is limited to trusted testers. Google plans access for AI Ultra subscribers through Project Genie and early partners through the Gemini API and AI Studio, without a firm date.

AI-generated summary, reviewed by an editor. More on our AI guidelines.

Google Launches Playground, Letting US Adults Build Browser Games From AI Prompts
Google launched Playground on Wednesday, an experimental browser platform that builds playable games from text prompts, open to US users 18 and older at playground.google. Its conversational interface
The Implicator
Repo Radar: 5 GitHub Projects Worth Your Week
Five projects climbed GitHub this week and none of them is a model. Each sits in the layer a team can own outright: the rules an agent reads before it types, the diagrams it hands back, the workspace
The Implicator
Apple Trains China-Specific AI Model With Alibaba, Sources Say
Apple has trained a large language model specifically for China, according to three people familiar with the unannounced work, Reuters reported in an exclusive. Alibaba helped train the model, while A
The Implicator
AI News

San Francisco

Editor-in-Chief and founder of Implicator.ai. Former ARD correspondent and senior broadcast journalist with 10+ years covering tech. Writes daily briefings on policy and market developments. Based in San Francisco. E-mail: editor@implicator.ai