I've had fun ragging on Tensor AI, Leonardo AI and MageSpace for their blunders, but lemme tall ya... Civitai can blunder an image just as badly as the others. For instance, this image. It is an image to image shot, which means it's meant to make a copy of the original image but gussy it up AI-style. Here's the image it was copied from:
This is a screenshot from Second Life using SL avatars and as you can see, the Civitai version is not an exact reproduction. What I wanted was an image of this semi-happy couple (the Slave Leia avatar appears to be scowling, and I love that) in more human, less cartoonish form. What I got was the two avatars merged into a single human form, the Civitai model I was using apparently not trained for fucking, a fairly common problem on Civitai. So it deleted the woman's hips and made the man's legs her legs. Well, fair enough, I guess. But then it twisted that poor avatar's back 180 degrees and also made it do a kind of loop. I actually think from the way her brawny shoulder and arm are positioned that she is supposed to be fisting herself from behind.
The AI also transformed her very nice panel gag into a crappily done cloth gag. But it did give her a humanish patina. Problem is, when you've done what the AI did to that avatar's body, no amount of patina is gonna help.
I did get better results eventually, though not perfect by any means.
