yanyuelouSDXL（伪）--基于kim大佬训练集训练

个人申明：

这个SDXL基于kim大佬的嫣月楼sd1.5的训练集训练的，

大佬发训练集到群里让我们练手的，想要的可以加大佬的qq群：852450267，kim大佬业务咨询 QQ95893026

训练底模和出图底模来自于我的base训练，我增加了一些自己的训练集想改变一下，但是由于自己失误，自己的图没生效，后续版本在重新修正下。出图在base和juggernautXL_v7Rundiffusion。

采样方法(Sampler)： Euler a | DPM++ 3M SDE Karras

放大算法：4x-UltraSharp

采样迭代步数(Steps)：20-30

高清修复采样次数：0-10

重绘幅度(Denoising):0.4

宽度 1024，高度 1024

lora权重0.75-0.95

感谢kim大佬的无私分享，感谢DukeG大佬的技术支持和指导。

Personal statement:

This SDXL is trained based on the training set of kim's Yanyue Tower sd1.5.

The big guy sent a training set to the group for us to practice with. If you want it, you can add the big guy's qq group: 852450267,

and the big guy kim's business consulting qq: 95893026

The training base and the output base come from my base training.

I added some of my own training sets to make changes, but due to my own mistakes, my images didn't work.

The subsequent version will be revised.The output is based on base and juggernautXL_v7Rundiffusion.

Sampling method (Sampler):

Euler a | DPM++ 3M SDE Karras

Zoom algorithm: 4x-UltraSharp

Sampling iteration steps (Steps): 20-30

HD repair sampling times: 0-10

Redraw amplitude (Denoising): 0.4

Width 1024, height 1024

Lora weight 0.75-0.95

Thank you for Kim's selfless sharing and DukeG's technical support and guidance.

Description