Civitaiトレーニング製

スカートめくれ。
元々、スカートめく「り」としてはモデルにタグがあるが、目的としてはスカートが腰辺りの何かしらに引っかかって、そのままめくれあがってる状態。
→Lora外しても出るが、このLoraではスカートの一部引っ掛かりを強調してる。
　全身描写だと発生確率、そこまで高くないが

スカートめくりのタグあるから軽度補正で行けるかと作る前は思ったがそうでもなかった。

■推奨強度
0.8-1.0　スカートの一部だけが固定的に上にめくれあがってるのが理想だが発生率が低い。
元々の学習画像が陰影強いお尻だけの部分使ってたせいで画風影響割と強め
基本的にAdetailer推奨

■その他
タグのいずれかが反応してるのか、全身描写側で画面２分割描写が出やすい。
ロングスカート類は構造的に無理が出やすいかな

■変な使い方
close-up でお尻を堪能できるゾ！

Trained on Civitai

This LoRA focuses on the concept of a flipped-up skirt.

While the base model may already recognize the tag for “skirt flip” (skirt lifted), this LoRA is aimed specifically at situations where the skirt gets caught on something around the waist, leaving it flipped up and partially stuck.

Originally, I thought a mild corrective LoRA would be enough due to existing tags—but that turned out to be insufficient.
→Even without the LoRA, it can still appear, but this LoRA emphasizes the caught/skirt-lift detail.

The occurrence rate isn’t that high when generating full-body shots.

■ Recommended Strength

0.8–1.0

The ideal output is where only part of the skirt is flipped upward in a fixed, natural-looking way, but the occurrence rate is low.

Due to the original training images using strongly shaded close-ups of the buttocks, the LoRA has a fairly strong visual style influence.

Using ADetailer is highly recommended for quality and structure.

■ Additional Notes

One of the tags might be triggering it, but full-body generations tend to result in split-screen-like compositions.
Long skirts tend to cause structural issues more easily.

■ (Silly) Alternate Use Case

Zoom in and enjoy the close-up of the butt! close-up is your friend.