Experimental SLIDER LoRA

Does NOT work very well with other loras.

And after a lot of on-site testing i found out that if the prompt say "Photorealist", "skin texture" etc this will reduce the quality and result a lot.
But at the same time if the prompt say "anime woman", "cartoon look" etc it helps a lot.

This LoRA is the result of a late-night experiment. This is a FUN lora collection, and dont expect any of them to be perfect or even good at every scenario.

• Variant 1: good balance

• Variant 2: Very good on female only, ok balance

• Variant 3: Very good female, not so good at male

• Variant 4: The worst in my testing

• Variant 5: Very good female, bad at male

While editing images in Klein, I replaced one person in an image with an anime character and started wondering whether I could train a slider LoRA to do something similar directly in zImage. That idea turned into this pack.

Getting consistent results was extremely difficult. In the end, I was left with five models that all worked reasonably well, but each of them developed its own style and strengths.

These LoRAs were trained quite aggressively, so they may noticeably change the original image composition even at moderate strengths. And some times it does not work at all.

From my testing, they seem to work best in the 5 to 6 range, both on the positive and negative side:

• Negative values makes the male characters more toward an anime look

• Positive values makes the female characters more toward an anime look

And in general, higher values tend to make heads abnormaly large, while lower values lean towards making the character look more "3D"

Just to be clear, I do not have any real knowledge of anime substyles, so some of these models may lean toward specific known styles without me being able to accurately label them.