This is a brilliant idea.
I'm sure what you're proposing isn't an easy turnkey solution but I'd love to find out how you do it. One of the difficult things, I'd imagine, is replicating Axl's incredible vocal range. I'd see if you can train or combine models on a few of his octaves - speaking simply "It's So Easy" vs. "Estranged" vs. "SCOM" - and be able to layer this into one track.
These AI cuts have been fantastic particularly where he could've used more time in the studio and the vocals lacked range/rasp and timing (TWAT first half, IRS, Catcher, Atlas Shrugged, State of Grace) vs. stronger vocal cuts (If the World, Better, Scraped, Eye on You etc) that sound like they were recorded before/after he had some vocal training.
Songs I'd love to hear vocals on are the instrumentals that I usually go back to:
Tonto
Mustache
Zodiac 13
Looking forward to seeing what you come up with!