V1.0
Strength recommendation:
0.7~0.9
Prompt:
gu1g58h
lora权重只要不是1,或者太低,都可以自行调试,因为在简单动作或者无表情的时候,权重为1,面部修复后五官的位置很奇怪,不过也并非不能使用1的权重,如果在面部没有出问题的情况下,可以尝试加深风格的 效果。手部如果出现错误的问题,请适当降低权重尝试,或者用其他lora修正,单独的lora 很难顾及到所有,为了保证已经得到风格不出现变动,不打算再去解决这个问题了
As long as the LORA weight is not 1 or too low, it can be adjusted manually. This is because when performing simple actions or showing no expression, the weight is 1, and the positions of the facial features after facial restoration are very strange. However, it is not impossible to use a weight of 1. If there are no issues with the face, you can try to enhance the style effect. If there are errors in the hands, you can appropriately reduce the weight or use other LORA corrections. A single LORA is difficult to cover all aspects. To ensure that the style has not changed and to avoid further solving this problem, we do not plan to address it anymore.
Description
FAQ
Comments (5)
Lucky that I searched for 'guigu' now.. there was only SD1.5 model until today morning. thank you 大哥!
btw, does this lora force to generate the figure in fixed pose or camera angle? I can see all your examples are head to legs body shot. Is it hard to generate full-length shot including shoes/feet?
@willshawn2519 It can generate full-body images, and you can also change the actions. However, I haven't included the relevant images yet.
兄弟,数据集有链接吗,我想制作一个z模型的版本
@pushpibl 数据我上传了你要是需要就直接下载吧














