How do you know that these people replace harassment with these pictures? And not just do both, or even increase their fetishes?

AFAIK psychologists simply don't know, and it might be a case by case thing.

What about the girls who’s pictures were used as material for these generated images?

As I explained, it might not be necessary to have any underage material in the training data.

Generally speaking I didn't come here to have a deep discussion about a very difficult moral and legal issue, I'll leave that up to the specialists. I wanted to say something about AI and somehow all answers I get are about the last tacked-on paragraph making a quick statement about me preferring keeping paedophiles away from kids.