This is more closer to actually creating rather than generating music. However this cannot be done with a text prompt, which the comment above claimed, is expressive enough.
Actually having an "autotune" AI that turns out-of-key poor singing into a beautiful melody while keeping the voice timbre, would be not bad.
