這個版本用了30張遊戲截圖，其中10張重複共40張。
推薦權重1.0~1.2之間，衣服裙襬與臉部特徵才能比較好還原。
這個LoRA的特色就是盡可能還原遊戲內的風格，由於資料集所有圖片都是去背，不打背景很大機率出現全白背景
This version was trained using 30 game screenshots (with 10 duplicates, totaling 40 images).
A recommended weight of 1.0–1.2 ensures the best reproduction of the skirt's hem and facial features.
This LoRA is designed to faithfully reproduce the in-game art style; since all images in the training dataset have had their backgrounds removed, omitting background prompts will very likely result in a plain white background.
