推荐参数：recommended parameters:<ul><li>提示词引导系数：5~6</li><li>迭代步数：20~30</li><li>采样器：DPM++ 2M SDE或DPM++ 2M</li><li>调度策略：Exponential或Karras</li><li>cfg：5～6</li><li>steps：20～30</li><li>sampler：DPM++ 2M SDE or DPM++ 2M</li><li>sheduling strategy：Exponential or Karras</li></ul>v1.1：<ul><li>支持24种不同风格</li><li>如果风格表现不明显，可以增大模型权重，推荐不超过1.6</li><li>基于kohakuxl-zeta-rev1训练，搭配kohakuxl-zeta-rev1出图效果更好</li><li>supports 24 different styles</li><li>If the style is not obvious, you can increase the weight of the model, and it is recommended not to exceed 1.6</li><li>training based on kohakuxl-zeta-rev1, so it works better with kohakuxl-zeta-rev1</li></ul>v1.0：<ul><li>支持25种不同风格</li><li>如果风格表现不明显，可以增大模型权重，推荐不超过1.4</li><li>基于kohakuxl-epsilon-rev2训练，搭配kohakuxl-epsilon-rev2出图效果更好</li><li>supports 25 different styles</li><li>If the style is not obvious, you can increase the weight of the model, and it is recommended not to exceed 1.4</li><li>training based on kohakuxl-epsilon-rev2, so it works better with kohakuxl-epsilon-rev2</li></ul>

multiple styles-pomegranate[sdxl]

继续训练v1.0得到v1.1，单个风格表现更明显，但部分风格间的区别仍较小，推测由不同风格数据混杂、数据标注不准确、模型不够拟合导致，后续版本将着重解决这些问题continue to train v1.0 to get v1.1, and the performance of a single style is more obvious, but the difference between some styles is still small, which is speculated to be caused by the mixed data of different styles, inaccurate data annotation, and insufficient model fitting, the subsequent version will focus on solving these problems

Files with SHA256: 17af3b00611782c739dec5159ad0795c56e29b0ce6509a6793eb3ba616130db6