[Update] あまりにも彩色や顔の再現度が低レベルだったので、作り直しました。今回から他のユーザさんを見習って network dimension (rank) を 8 としてサイズを減らしています。
==========
This LoRA has "v2" in the file name, but it is not a failure. It is trained twice with 8 images, 10 epochs, 10 iterations for 200 steps, and more with 20 images by the first LoRA, 10 epochs, 5 iterations for 250 steps. Trigger and useful keywords are as follows.
Appearance: sdxl-cowboybepop-julia, light brown hair, straight hair, hair between eyes, long hair, green eyes, makeup, lipstick, mature female
かなりの暴れ馬です。たぶん50枚に１枚くらいしか本当に似てるの出ないかも。実際、サンプルの最初の画像を得るまでに200枚出しました。
Description
Trained completely new by 20 images (restored with Nano Banana), 10 epochs, and 2-10 repeats for 330 steps.
Useful triggers and tags (keywords) are as follows.
Appearance: sdxl-cowboybepop-julia, blonde hair, straight hair, hair between eyes, very long hair, grey eyes, makeup, lipstick, mature female
Outfit: black clothes, black coat, long coat, black turtleneck, black boots