Google 發佈 Genie 4 世界模擬模型，先向指定合作夥伴開放研究預覽

Google is releasing Genie 4 today, a new world simulation model that can create photorealistic videos with ambient noise and dialogue. The new model is launching first as a research preview for trusted partners, but it will soon be available to Google AI Ultra subscribers via Project Genie, a web app where users can generate interactive environments they can explore like a video game.

I wrote about Project Genie back in January when the experimental project was leveraging Google’s previous Genie 3 model. At the time, Google AI Ultra subscribers could only create worlds that could be explored for 60 seconds, but the new Genie 4 model will bump that limit to 2 minutes. Moreover, Genie 4 makes it possible to modify a simulation in real time using a natural language prompt.

“With Genie 4, you can set an event in motion and steer scenes in real time using natural language, like moving the camera, changing the weather, or introducing new characters. Every world also generates native audio on the fly, including ambient sounds, music, and interactive dialogue,” Google explained today.

Genie 4 uses Google’s Gemini models behind the scenes, including the new flagship Argon model, to model multi-step events and keep simulated worlds coherent. While it’s still early days for real-time world models, Genie 4 is Google’s first model that can generate interactive audio on the fly. However, Google acknowledged that “visual details and object permanence can still shift or drift over time.”

If you’re a Google AI Ultra subscriber, you can join the waitlist to get access to Genie 4 in Project Genie here.