succubus (Vindictus) LORA

测试 LORA

---

提供者：Buddy17

"Vindictus 中的魅魔 + testfield"

/model/205727/succubus-from-vindictus-testfield

---

请添加您喜欢的姿势、背景、服装等。

Add your favorite poses, backgrounds, costumes, etc.

---

权重 0.5 ~ 0.7，使用 ADetailer 并通过 face_yolov8n 进行面部修正。

高分辨率修正：R-ESRGAN 4x+ Anime6B，放大倍数 2 倍

Weight 0.5 to 0.7 ADetailer is used and face_yolov8n is used for face correction.

Hires.fix: R-ESRGAN 4x+ Anime6B, 2x magnification

---

印象

在原始数据中加入了两张 AI 输出图像进行训练。

我认为面部和身体的数据已经足够。

服装的还原稍显模糊，或许再增加 2-3 张图像会更好。

大镰刀难以训练，我认为需要更多教师图像。

手部形态非常难以生成，因此需要更多手部数据。

Learn by adding two AI output images to the original data.

I think the data on the face and body is sufficient.

The reproduction of the costume is a little vague, so it might be nice to have two or three more.

Scythe is difficult to learn, so I think more teacher images are needed.

I think we need more data on the hand because it is very difficult to form the hand.