Try size: 1024 x 1024 (anima and ILXL) / 640 x 640 (SD1.5)
Weights: 1.0
Clip skip: 2
Training material from bing and game screenshot
Note: This is my most recommended version! The advantage is that the training results for the female character D·Yuri·Anna are quite stable; the disadvantage is that the hair ends of the anime characters have gradients and splits, which have been completely incorporated into the model training. / 這是我最推薦的版本！優點是女性角色D·Yuri·Anna的訓練成果算是相當穩定；缺點就是動漫人物的髮尾有漸層跟分岔，完全被模型訓練進去了。(Anima)
Note: It's impossible to train the background of de_dust2 into Lora, but illustrious is better at character creation. / 將de_dust2的背景訓練進Lora是沒辦法做到的，但illustrious比較擅長角色塑造。(ILXL)
Note: Because this is a very early version of LoRa, I do not recommend using this version. / 因為是很早期訓練的Lora，所以我不建議使用這個版本。(SD1.5)