Karma_Luey版本1.0 (ID: 985427)

This is the same concept as the Hyoon model, to save space here, I recommend just reading that one to understand how this one works https://civitai.com/models/912708/hyoonaikurosfloaromaa-1cfg-face-edition . TLDR: Use 1 cfg for the inpaint/adetailer. Use 0.75 weight 4-9cfg for harder main prompts.

There are some funky habits with this Lora because Karma doesn't post many high quality pictures, so a lot of it is screenshots from her reels. In a lot of them, she squints her eyes while smiling, so the default is to have squinty eyes. So I suggest using "eyes closed" in negative to prevent this.

Most of the pictures are also very low quality, so by default the eyes are blurry, but eyes closed will also remove this because I included some higher quality pictures where her eyes are wide open so the lora will generate more clear eyes if "eyes closed" is in negative.

It will also freak out if you use hair/eye colors that's not in the training data which is just, red, blonde, black hair and brown, blue eyes. I do intend to fix this by photoshopping the eyes and hair on a couple images to be more variety of colors. But it is a lot of work so we will see if I can muster up the motivation to do it. But within those constraints because of her lack of high quality images, the Lora is quite good.

描述:

This is 20 repeat 1 epoch, undertrained so I don't use it anymore

训练词语: Karmaluey