>>718441153
I do understand it to some extent though.
Take this pic for example.

I usually only have information extracted between 0.20 and 0.45
When on that lower end, it transforms the pic to be more lineart focused.
If I dial it up, the colors become more watercolory.
"Wow that looks really nice. Now, what happens when I go further than that again..?"
Well, then, it becomes DETERMINED to have the characters lying down like this kid is here.
And I guess every pic is gonna have a weird tradeoff like that one.

>>718258139
Here is the vibe reference in question.

Reference Strength: 0.60
Information Extracted: 0.19

I dunno which models outside of NAI even use external images in this way. But in case none of them do: It's NAI.

Whoa.

NAI users, try this image as a vibe ref, with relatively high reference strength (0.6-0.9) and low information extracted (0.15-0.30)