。这是我练的一个LORA，采集来源KS，训练图片23张，尺寸768*768，图片数量*10*30/*2。该模型泛用性极高，适用各种大模型。DPM++ 2M Karras 20-28。建议0.5~0.8。0.5~0.8建议开启ADetailer。0.5泛化性高加上修复脸部给到0.8权重+ADetailer修复。高清采用4x_NMKD-Siax_200k，10步，0.25-0.4。本人炼制真人模型有一定的经验可以还原素材80%-95%。接受真人定制（本人只练真人，其他一概不懂。。。）。如需要数据集/训练参数，请联系本网站私聊。更多福利可以移步CCCCC，或联系本人。发布的LORA请多多返图，感谢支出。出品不易，欢迎讨论，学习，交流。QQ：39826676（福利）. This is a LORA I practiced, collected from KS, 23 training pictures, size 768*768, Number of pictures1030/*2This model is extremely versatile and suitable for various large models. DPM++ 2M Karas 20-28. Recommend 0.5~0.8. 0.5~0.8 It is recommended to open ADetailer0.5 High generalization plus repair and repair gives a weight of 0.8 + ADetailer repair. HD uses 4x_NMKD-Siax_200k, 10 steps, 0.25-0.4I have some experience in refining real-life models and can restore 80%-95% of the original materials.. Accept customization by real people (I only practice real people and know nothing else...). If you need data set/training parameters, please contact this website for private chatFor more benefits, please go to CCCCC, or contact me. Please return more pictures after publishing LORA, thank you for your expenditure. It’s not easy to produce, welcome to discuss, learn and communicate.QQ：39826676（Welfare）

Files with SHA256: fb3ed94ad6211e112a513c826c647d0f39555e1ed6c2db49dafd1958815313ed