A LoRA that generates panties that are caught on one leg when they are about to be taken off.
Ver.3 with higher performance was completed because the understanding of LoRA has improved a little.
Compared with Ver.2,
The impact on image quality has been greatly suppressed, and the probability of generating panties has also been improved.
Be sure to use the "panties around one leg" prompt.
Up to Ver.2, the recommended strength was 0.6 to 0.8, but it seems that many models of Ver.3 do not distort the image even with 1.
It works well with 2D and 2.5D illustration models, but it seems that it is not good at live-action models, and the probability drops considerably.
others prompt,
"bottomless"
"skirt lift"
You may add more as needed.
I'm still dissatisfied with the shape of the generated panties, so if my understanding of LoRA improves and I can complete a good LoRA, I'll publish it, but I have no idea when that will be. I do not understand.
(May 8, 2023)
Added Ver.2.0. See "About this version" for details.
(June 2, 2023)
Added Ver.3.0.
///以下翻訳前の原文///
脱ぎかけで片脚にひっかかってるパンツを生成するLoRAです。
LoRAへの理解が少し向上した為に、より性能の高いVer.3が出来上がりました。
Ver.2と比較すると、
画質への影響が大きく抑えられ、またパンツの生成確率も向上しました。
"panties around one leg" promptを必ず使ってください。
Ver.2まではお勧めの強度は0.6~0.8まででしたが、Ver.3は1でも画像が乱れないモデルが多い様です。
2次元、2.5次元イラストのモデルでは上手く動作しますが、実写のモデルは苦手の様で、かなり確率が下がります。
その他、promptに
bottomless
skirt lift
などを必要に応じて追加しても良いでしょう。
まだ生成されるパンツの形状に不満があるので、もしLoRAへの理解度が向上し、良い出来のLoRAが完成する事があれば公開しようと思っていますが、それがいつになるかは全く解りません。
Description
Since my understanding of LoRA has deepened a little, I made ver.2.0.
Compared to ver.1.0, the probability of creating objects that look more like panties has increased.
However, if you don't use it with a strength weaker than ver.1.0, a lot of panties will be generated in one picture, and a strange human body structure picture will be generated easily.
It seems that the optimum intensity differs depending on the model, so I would like the user to find the optimum while adjusting it himself.
My quest for panties seems to continue.