两年没使用过AI了，突发发现现在AI绘图的质量有着巨幅提升，所以重新训练了罗小黑人物的Lora。

It's been two years since I last used AI, and I suddenly noticed that the quality of AI-generated artwork has improved dramatically. As a result, I retrained the Lora model for the character Luo Xiaohei.

使用WAI-ANI-NSFW-PONYXL v12

模型进行训练，推理目前只测试了训练的模型。

虽然在使用时发现这个检查点已经有点过拟合了（发现打标也有一点小问题），但是目前我不想继续花费跟多花间去重新训练寻找一个更好的检查点。

以下是一个简单的使用说明：

Prompt:

score_9, score_8_up, score_7_up, source_furry, luoxiaohei, child,
// 加上你想要的具体描述 （建议加上 cat boy, cat ears, cat tail, green eyes, black hair）
<lora:reg_luo_xiaohei_(human)-000021:1>

Negative:

score_6, score_5, score_4, source_cartoon, unknown object, 3d, censor, monochrome, blurry, lowres, watermark

Width/Height: 1024x1368 （发现这个分辨率似乎更稳定）

Guidance scale: 6~7

Steps: 30±

Sampling method: Ruler a

Sampler Eta: 0.67

Clip Skip: 1

Model: waiANINSFWPONYXL_v12

VAE: None

Using the WAI-ANI-NSFW-PONYXL v12 Model for Training

Inference has only been tested on the trained model so far. Although it was found that this checkpoint is somewhat overfitted (with some minor tagging issues), I currently do not want to spend more time retraining to find a better checkpoint.

Below is a simple usage guide:

Prompt:

score_9, score_8_up, score_7_up, source_furry, luoxiaohei, child,
// Add your specific descriptions here (suggested: cat ears, cat tail, green eyes, black hair)
<lora:reg_luo_xiaohei_(human)-000021:1>

Negative Prompt:

score_6, score_5, score_4, source_cartoon, unknown object, 3d, censor, monochrome, blurry, lowres, watermark

Settings:

Width/Height: 1024x1368 (found that this resolution seems more stable)
Guidance scale: 6~7
Steps: 30±
Sampling method: Ruler a
Sampler Eta: 0.67
Clip Skip: 1
Model: waiANINSFWPONYXL_v12
VAE: None