This "fine-tune" is broken
Yes, it gets rid of the "AI look", but It suffers from some of the things I would see in an amateur SDXL fine-tune: "same pose syndrome" - the model no longer understands different shot types well (portrait, upper body, full body, etc); more body horror - how you managed this is beyond me. Similarly, more difficulty with understanding what clothing looks like; biased towards Asians: most of the time if I prompt "a pretty brunette woman" I'm going to end up with the same or similar looking person.
I may be a jerk, but I'm also not wrong. This is trash.
seems fine to me
seems fine to me
LOL
finetuning is finetuning.
I agree, I'm using an FP8 model, maybe it's a bad conversion but for me it's softens the impact of the generation, lots of flux 1 dev loras don't work well with it, it's like a watered down version to try and eliminate the chin taste but end up reducing the richness of the other flavours (colours/ dynamic contrast). Maybe I'm using it wrong, or haven't set it up right?