and i'm building things with musicgen right now with one simple idea:
removing the text from ai music. musicgen specifically, is more like a little musician itself than other models, and musicians shouldn't have to describe what they want from each other. they jam.
using your audio as input, gary comes in with the fills.
suddenly, you got yourself a lil jam buddy.
max4live device and chrome extension coming, but on hf you can already play with a version that has dope UX thanks to gradio, and is free thanks to zero-gpu-explorers.