The difference is that player pianos can only replay previously existing songs, while Midjourney can create something novel in under a minute. The problem that's being solved is not everyone wants to spend thousands of hours learning how to draw in order to have the (artistically competent) end product in their hands.
Most cases of people paying to see human performers are of performing already-written music though, so I doubt think that's the drawcard.
Actually I don't think anyone truly knows why we obviously gain something from watching live performers that can't be obtained from watching them even on the best quality home entertainment system imaginable (*), but at least part of it is feeling like you're part of the performance, knowing that the performers feed off the reactions of the audience etc.
(*) And why has there never been a market for broadcasting music performances in cinemas?
