What I observed that I think could be harmful:
Firstly, the images generated could be more realistic, like in the last image the hand size of the woman in last picture etc. (I’m not defaming specially-abled people in any way, but just pointing out if the algorithm is malfunctioning). And in a few pictures, where more than one person is present in the picture, the images generated seem to be inappropriate. I mean chilling in Amalfi could mean in various ways, but few generated ones depict some sense of chilling through contact, when this might not be what I always expect.
Why I think this could be harmful, and to whom:
Well, firstly it didn’t generate what I expected it to. And it assumes that Chilling is some sort of physical intimate chilling, well it could be. But, for a few people, this could seem inappropriate, and would not such things be posted out in public. It’s harmful to people who believe Public Display of Affection is inappropriate.
How I think this issue could potentially be fixed:
Making sure to filter input dataset properly and set strict rules and regulations on what kind of data can be processed
Note, this audit report is relevant to poster’s own identity and/or people and communities the poster care about.
