<h1 id="slender-+-big-breasts-win">Slender + Big Breasts = Win</h1>This LoRA is designed to create a slender physique with large breasts.Since I used masks for the background, face, and hair during training, it should have minimal impact on those elements. However, because it significantly alters proportions (such as head size and limb length), it may affect the Krea2 model's inherent adjustments—for example, shifting facial features from a standard body type to a slender one.I also included clothed data during training, so it should work reasonably well even when the character is wearing clothes.In my personal opinion, at a LoRA strength of around 0.7, the character should remain recognizable (i.e., "This is what they would look like if they were thinner").Please note that if your pose prompts are ambiguous, the pose might change a bit due to the changes in proportions and head-to-body ratio.<h1 id="+">スレンダー + 巨乳 = 優勝</h1>このLoRAはスレンダーボディに大きな胸を持つ体型LoRAです。 背景と顔と髪にマスクをかけた状態で学習させたので 体型以外の要素には基本的に影響がほぼ無いはずですが、 頭身やパーツの長さ等が結構変わるので、Krea2モデル本来が持つ調整 例えば標準体型に準じた顔の造形から痩せ型体型に準じた顔の造形に変わる などの影響は受けます。 学習の際に着衣データも混ぜたので衣服を着た状態でもある程度機能すると思います。個人的な主観ですがLoRA強度0.7ぐらいまでなら まぁこの人が痩せたらこんな顔になるのかなぁ と一応同一人物認定できる程度にはなってると思います。またポーズに関してプロンプトの記述が曖昧な場合は 頭身の変化などでポーズが結構変わってしまう場合があります。

Files with SHA256: b1d91b954b49d8cf550a451474d45595a3d7d028682247d9b743a66bfee00766