I gotta go contemplate 'where we're at' again it seems. If that is truly a straight generative audio diffusion model.... wait, how did they get the same verse by verse chord progressions to match? this has to be professionally post-produced, right? AI models aren't able to do this end-to-end yet, right?
[0] https://www.youtube.com/watch?v=_88Qg8FGrqY [1] https://www.youtube.com/watch?v=8gFKREP3gPg
I'm ashamed to say I prefer these to the originals so much so that its difficult for me to listen to them any more. Make of that what you will...
Usually these AI covers don't use AI for the whole thing, but rather specifically for melding the to-be-impersonated voice into some given melody. That's been possible for a couple years now with decent results; one of my favorite examples is that of Plankton from Spongebob singing Disturbed's cover of “Sound of Silence”: https://www.youtube.com/watch?v=7eLRsw9mkmY
Possible that these newer ones are also using AI to generate other musical elements, but it's probably all being combined after-the-fact rather than being generated all at once.