Who can try Google's Genie 4 world model?
Google has opened Genie 4, its newest AI model for generating interactive worlds, to paying subscribers. The company announced the rollout on Wednesday. Eligible Google AI Ultra subscribers can now try the model in Project Genie, a browser-based prototype. Google AI Pro members can join a waitlist.
Genie 4 creates a world that users can move through, and it lets them change that world as they go. Typing a prompt mid-session can shift the weather, add a character or redirect what happens next. The model also adds sound for the first time, covering everything from background noise to character dialogue.
What new features does Genie 4 add?
Google DeepMind first showed Genie 4 to testers in October. The public release adds two features that were not part of that preview. The first is an AI agent that builds worlds from a conversation with the user. The second is a gallery where people can share their creations and remix each other's worlds.
Project Genie already lets some users explore simulated cities built from Google's Street View imagery. That gives the prototype a practical hook, since people can walk through familiar streets and then alter them with a few words.
Google is not pulling back its older model. Genie 3 remains available for longer runs. Each generation with Genie 3 now lasts up to three minutes, up from 60 seconds.
The technology has drawn attention from the video game industry, where many big studios remain wary of generative AI. Take-Two Interactive said that Grand Theft Auto 6 did not use any generative AI. Google's own materials make clear that the company is not overselling what Genie can do. The company says both Genie 4 and Genie 3 are still research systems. Details can drift during long sessions, and readable text and complex physics remain hard for the models to handle.
Google is also marking the output. Everything created in Project Genie carries SynthID, the company's invisible watermark for AI-generated content.
Investors are paying close attention to world models. In September, AMD agreed to buy World Labs, the startup led by Fei-Fei Li, for $8.2 billion. Google DeepMind has argued that such models could help train robots and AI agents before those systems are deployed in the real world. Odyssey, a rival lab, says its Odyssey-3 model can already control robots and cars.
Is Genie 4 still experimental?
For now, the Genie 4 rollout is limited to a small group of paying customers and waitlisted users. Google has not said when access will widen beyond those groups. The company's own warnings about accuracy suggest that the model is still in an experimental stage, even as it moves into the hands of paying users.
Questo articolo è stato scritto con l'assistenza dell'IA.
News Factory APP - notizie agentiche per potenziare il tuo SEO e AEO.
