Original title: Breaking | Google just unveiled Genie 4! First to support native audio and improvised dialogue
Author: Wanshan Yibasuoluo
Source: Duanku Zixishi (Short Drama Study Room)
On October 28, Google DeepMind officially released Genie 4, its next-generation general world model.
With a single prompt, it can directly create a living universe that can respond in conversation and change course in real time.
Google CEO Sundar Pichai announced it immediately: from generating a video clip to running a world in real time, the oasis era of "words become reality" has arrived. He also called Genie 4 Google's most realistic and interactive world model to date.
Here's the key point: this time, Genie 4 has pushed the world model into a brand-new stage with sound, people, and a brain.
It supports native audio generation for the first time, synchronously producing ambient sound and background music, and enabling characters to improvise dialogue as scenes change. Users can also alter events, camera angles, and actions at any time to steer dynamic environments in real time.
In terms of interaction and memory, a single coherent world generated by the model can last about two minutes at roughly 24 frames per second. Meanwhile, micro-expressions, mirror reflections, and physical collisions have all been improved, bringing the virtual world closer to real environments.
Google also demonstrated the ability of multiple Gemini models to work together, covering scene understanding, complex action planning, and robot task execution.
Unlike models focused on generating a few seconds of video, Genie 4 aims to build an interactive world that responds to instructions in real time and keeps evolving. Such technology is expected to provide a virtual environment for general agent training and robot simulation that can be iterated through trial and error.
Genie 4 is now in closed beta, and Google AI Ultra users can apply to join the waitlist.
