Version 3.0 is finally.
Training size 512*768.
The success rate has been greatly improved.
<lora:cover eyes3.0:1>
After testing more than 20 models, they will basically be successful, and you can try other models.
Here are the models I recommend using with a high success rate:
onlyrealisticV30.TlkW
brav7finalFp16201.uDDf
counterfeitV30Fp16.UKAC
yutamix20aqua.QkO6
dreamshaper_5BakedVae
darkSushiMixMix_brighterPruned
3.0版本终于来了。
训练尺寸512*768。
此版本大大提高了成功率。
<lora:cover eyes3.0:1>
测试了二十几个模型，基本都会成功，可以试试其他模型。
以下是我建议使用的模型，成功率高：
onlyrealisticV30.TlkW
brav7finalFp16201.uDDf
counterfeitV30Fp16.UKAC
yutamix20aqua.QkO6
dreamshaper_5BakedVae
darkSushiMixMix_brighterPruned
Description
It can work on real scenes, but there is still a probability of failure, I guess it is a problem with detecting faces, but I don't know much. But it works.
【Note】Negative add : blindfolds, masks
and CFG Scale set 12+
can Increase success rates.
Thank you for liking it, please give me 5 star!
2.0版本可以用于现实场景，比1.0效果好，二次元的表现也比之前的效果好了。
【注意】负面提示加上：blindfolds, masks
相关性调到12+ 在现实场景可以提高成功率。
但是还是有不如意的地方，我猜测是人脸识别的问题，当然我只是猜测，因为我一窍不通。
谢谢你的喜欢，给我个五星好评吧！