Brief IA
›

Google Makes Genie 4 Available Through Its AI Ultra Subscription

Models & LLM·Tom Levy·

Google Makes Genie 4 Available Through Its AI Ultra Subscription

Google Makes Genie 4 Available Through Its AI Ultra Subscription
Key Takeaways
1Eligible subscribers to Google AI Ultra can try Genie 4 in Project Genie.
2Genie 3 remains available, with generations lasting up to three minutes.
3Google warns that long sessions may lose details.
Why it matters : Genie 4 allows for transforming a world during its exploration and adds native sound generation.

Google is granting access to Genie 4 to eligible subscribers of its paid offering Google AI Ultra, through Project Genie. This latest model for generating interactive worlds allows for modifications during exploration and now produces sound.

Google Keeps Its Models in Experimental Stage

Google continues to present Genie 3 and Genie 4 as research devices. The company warns that prolonged sessions may lead to a loss of detail; obtaining readable inscriptions or reproducing complex physics remains challenging.

For longer explorations, Genie 3 is still available: each generation can last up to three minutes, instead of the previous 60 seconds. All creations from Project Genie receive SynthID, an invisible watermark designed by Google for AI-generated content.

Project Genie Allows for World Remodelling

Introduced to testers by Google DeepMind in October, Genie 4 is used within Project Genie, a prototype accessible via a browser. AI Pro subscribers have a spot on a waiting list. Users can navigate the generated environment and enter instructions to change the weather, introduce a character, or alter events.

La lettre du soirGratuit · 19 h

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

The native sound generation, new for this model, covers ambient sounds as well as character dialogues. An agent can also create worlds through conversation, while a gallery allows users to showcase their creations or transform those of others. Some users can already explore urban simulations created from Street View.

DeepMind Also Considers Training Robots

The potential applications extend beyond gaming: Google DeepMind believes these environments could prepare robots or AI agents before they transition to the real world. The competing lab Odyssey claims that its model Odyssey-3 can already command robots as well as cars.

Major video game studios remain cautious: Take-Two states that it has not employed any generative AI for GTA 6. Investments are also directed towards these models: in September, AMD agreed to acquire World Labs, the company founded by Fei-Fei Li, for $8.2 billion.

Brief IA — L'actualité IA en français

L'essentiel de l'actualité de l'intelligence artificielle, décrypté et expliqué chaque jour.