It appears to be linked to certain style tokens. Like the image is being routed to a system that is weighted incorrectly? I know nothing about how this stuff works but I noticed that the issue was manifesting differently for different styles.
I gotcha, I work with huge prompts majority of the time. I was offering my first thoughts on a key token structure. Do you have a specific example you could share? I don’t mind doing some testing trying to achieve a better set of tokens to share.
I have a prompt for this (kind of);
(Character A’s Masculine Body Type) = an adult with maturely defined bodily proportions. With a Lean runners build. With Low body fat. With Wiry muscle mass. With Long limbs. With a Narrow frame. With wide feminine hips. With thick rounded thighs.
From here you’d add whatever else you’re looking for. Hope this helps!
If that’s not tall enough, just add “tall” between ’a’ and ‘ponytail’, it should work. If this isn’t what you’re looking for then I am sorry.
“Front view of Woman with her hair slicked back. You see a ponytail raised up from behind her head.”