是一只九龄 - Famous Chinese cosplayer Yizhijiuling

展示图里效果特别好的（比如校服生活照、暗背景写真）都被CivitAI给删除了，不知道他们到底什么逻辑。

v1.0 版本使用了全新的训练参数，结合原始数据集和增强的面部数据集进行多尺寸训练，并去除了绝大部分数据中的水印。现在可以更好地生成面部、全身，并组合其他LoRA而不至于让形体崩坏了。

The v1.0 version uses new training parameters, combines the original dataset and the enhanced face dataset for multi-size training, and removes watermarks from most of the data. Now it is possible to better generate faces, full bodies, and combine other LoRAs without causing the shape to be corrupted.

使用 200 张照片 5 epoch 训练的 LoRA，混合了九龄的 Cosplay 照和日常照
Leveraged 200 photos trained for 5 epochs, the training data is combined with her cosplay and daily-life photos

由于九龄的各个作品妆造差异较大，出图需要比较多的描述来获得稳定的结果。适合搭配提示词 young asian woman / cute teenage girl 等。

Due to the high variance of the make-ups across her works, detailed caption is needed for a stable result. Recommended to use with promts like "Young asian woman" or "cute teenage girl", etc

此版本部分训练数据未能去除水印，暂且明明为v0.1 版，水印和稳定性问题之后想办法修正。

This version has training data that didn't have watermarks removed, temporarily named as version v0.1. I will try in the future to deal with the problems with watermarks and stability issues.

推荐FLUX引导2.5-3.5，LORA权重0.9-1.1

Recomended FLUX guidance 2.5-3.5, LORA weight 0.9-1.1

To CIvitAI stuffs:

She IS an adult, stop deleting my posts with NORMAL school uniforms