Li Chunyan / Girl Friend (Beta) / 李春燕 / ガールフレンド(仮)

This little few images LoRA training purpose was learning LoRA specification deeper for me so probably I got more accurations. (この画像の少ない) LoRA トレーニングの目的は私がLoRA 仕様のより深い学習、恐らく正確な成果を得たと思います。)

---

Trained separately without regularization. (正則化せずに分けて学習させました。)

Trigger word: Additional spell if doesn't appears. 　li_chunyan: blue eyes, purple short hair, double hair bun, bun cover, (トリガー　：　出ない場合の追加の呪文)

Additional trigger: Additional spell if doesn't appears. 　seio_uniform: blazer, long sleeves, school uniform, plaid skirt, 　seio_summer: white shirt, plaid skirt, blue necktie, 　seio_buruma: gym uniform, white shirts, buruma, red buruma, (追加トリガー　：　出ない場合の追加の呪文)

Special trigger: Additional spell if doesn't appears. 　panda-toy: panda 　dimsum-toy: baozi (追加トリガー　：　出ない場合の追加の呪文)

Could dismiss real-size panda and non GF baozi.

*** Included (Puti?Puchi?)Petit-Girl-Chan, Atlas(MAP, appears similar chibi), Standing Picture Style(Experminal) (ぷちガールちゃん、地図キャラ、立ち絵を含めました)

I don't know that's efficient method. I could not enough tested through slowly cpu and gpu.

*Only prompt "petit_girl"(& other) and exclude "mirei" tag when wants petit girl(chibi). If doesn't so, mixed semi puti-girl figure appears within specified on both prompt. (ぷちガールちゃんを出す場合、呪文には「petit_girl」(やその他)のみとして、呪文から「li_chunyan」のタグは除外してください。 除外しない場合は混ざります)

🤔🤔🤔🤔🤔　　　🤯

Additional trigger: Additional spell if doesn't appears. 　petit_girl: chibi 　petit_atlas: chibi 　stand_pic: (Unknown) standing, full body? (追加トリガー　：　出ない場合の追加の呪文)

I can includes onpu_style and others(atlas...etc) on another definitions on the next time too if this effective.

So increasing studies count to double or triple(~5th) when since few image. (元画像が数枚しかないものは学習数を倍〜5倍程度に増やしています)