Moral and legal discussion aside, this is technically very impressive. I also wouldn’t be surprised if this somehow kickstarts open source music generative AI from China.

This already exists and is interesting to play around with - https://github.com/ASLP-lab/DiffRhythm