One take several exposures and average them together to reduce noise and increase dynamic range and to me that's still an adequate representation of reality. In fact, they're really just techniques used to overcome the limitations of today's sensor technology.

Blending multiple exposures of the same scene is still a representation of what the camera/sensor could see.

But when you start copying and pasting imagery from one photo onto another, then in my opinion you've crossed a line into fakery. One can say that it's simply adding detail that the camera can't see... but we're assuming that detail is in fact accurate.

Is it in fact an accurate presentation of the moon? Is it correctly showing the current phase? What about adding those same details to a streetlight?

And where does the AI substitution stop? If I take a photo of a car fleeing from a hit and run, is that really the license plate number? Or is it one the AI decided to add to simulate extra detail?