这是根据一组古灵精怪的小小姐照片练的lora，底模是麦菊大佬的majicmixRealistic_v6，训练的图片大小是512*768。lora权重0.8就够了。
正面提示词；masterpiece,best quality,1girl,solo,(Ultra-realistic 8k CG: 1.2), perfect artwork, exquisite patterns, intricate details, (unparalleled masterpiece, best quality: 1.2), (extremely complex: 1.2),<lora:caozhangyingfei:0.8>
负面提示词； bad-picture-chill-75v, bad_prompt_version2-neg, BadDream, badhandv4, By bad artist -neg, EasyNegative, EasyNegativeV2, FastNegativeV2, negative_hand-neg ng_deepnegative_v1_75t (1),(worst quality:2), (low quality:2), (normal quality:2),(uneven eyes),lowres, normal quality,bad anatomy,bad face,(uneven eyes),paintings,ugly, bad hands,open mouth,multiple girls,extra faces, extro breasts, multiple breasts,obese, fat rolls,extra arms, extra eyes,inverted nipples,extra ears,nipple rings,severed arm, bad arm, nipple bar,asymmetrical eyebrows,big mouth,.
迭代步数；25~30 采样方法；dpm++2m karras 提示词引导系数；7 大模型用写实系列的都可以。
Description
这个版本改变上一个版本的水印问题，我对之前的数据集进行了重新训练，现在的版本不会再出现水印了，正反面提示词和迭代步数按照上个版本的数据就可以。