先看说明

鹅群：

墨幽炼丹阁-太清阁：858495398

墨幽炼丹阁-上清阁：764481772

（群文件有含信息的参考图包）

也请多多关注我的小红书和抖音账号：yy469370733

非稳定模型

由768以上分辨率训练，出图分辨率可用更高分辨率：768×1152、768×768、576×853（亦可视情况自行调整尝试）

例图均为高分辨率+fix直出

基础设置推荐：

tag：

正：1girl,moyou(可选)

负：(low quality:2),

参数：

CLIP 终止层数：2

vae：已内置

采样：DPM++ 2M Karras（其他采样亦可自行尝试）

步数：20~50 （见上方采样推荐）

cfg：7~8

高分辨率修复参数：

高分迭代步数：0（0即与迭代步数相同）、或20

放大算法：4x-UltraSharp、

重绘幅度：0.2~0.3

基础分辨率：短边512~768

放大倍数：1.5~2

Let's take a look at the instructions first

Goose flock:

Moyou Alchemy Pavilion - Taiqing Pavilion: 858495398

Moyou Alchemy Pavilion - Shangqing Pavilion: 764481772

(The group file contains a reference image package with information)

Please also pay more attention to my Little Red Book and Tiktok account: yy469370733

Unstable model

Trained with a resolution of 768 or higher, the image resolution can be higher: 768 x 1152, 768 x 768, 576 x 853 (or adjust and try according to the situation)

Example images are all high-resolution+fix straight out

Recommended basic settings:

tag：

Positive: 1Girl, moyou (optional)

Negative: (low quality: 2),

Parameters:

CLIP termination layer: 2

Vae: Built in

Sampling: DPM++2M Karras (other sampling can also be attempted by oneself)

Steps: 20-50 (see sampling recommendation above)

cfg：7~8

High resolution repair parameters:

High score iteration steps: 0 (0 is the same as the iteration steps), or 20

Amplification algorithm: 4x UltraSharp

Redraw amplitude: 0.2~0.3

Basic resolution: short edge 512~768

Magnification factor: 1.5-2