tongliya 哎呀呀是丫丫呀 for SDXL1.0 & SD1.5 #yaya版本SD1.5 v1 (ID: 233973)

Please leave a review if you like this lora！

SD1.5

用同样的素材缩小，做了个SD1.5版，还行，还行,

Reduced the size using the same materials and created an SD1.5 version, which is okay, okay.

SDXL1.0 v2

可了不得，在网上收集，增加了很多原始素材，最后训练图片数量达到了250张，嗯对，二百五，Celebrity类的lora其实不需要这么多素材，可想想，反正玩玩也不错，于是用我的RTX3060显卡训练了15个小时，epoch=8，以后可能再也不这么干了，哈哈,

素材都是选择比较高的素质，同v1.0一样采用1152分辨率，不同的是，都是至少不小于这个分辨率的原始素材压缩而成的，酒窝也炼出来了，目前发上来的epoch=5的版本，dim=32，文件大了些，也没有压缩，使用确实比上一版本好一些，但仍建议使用ADetailer 可以得到更好的效果,

It's amazing, I collected a lot of raw materials online and added them. In the end, the number of training images reached 250. Hmm, two hundred and five. The Celebrity Lora doesn't actually need so many materials, but I think it's good to play anyway. So I trained for 15 hours with my RTX3060 graphics card, epoch=8, and I may not do it again in the future, haha.

The materials are all of high quality, using a 1152 resolution similar to v1.0. The difference is that they are all compressed from raw materials at least no less than this resolution, and dimples have also been refined. The current version with epoch=5, dim=32, has a larger file size and no compression, which is indeed better than the previous version. However, it is still recommended to use ADetailer to achieve better results.

SDXL1.0 v1

刚开始玩SD的时候，就炼了个SD1.5的丫丫，不过没有发出来，用同样的素材，1152的分辨率，制作了SDXL版本的lora，试了一些大模型的效果，发现需要ADetailer才更像，下一版本，打算加进一些新的素材，再看看效果,

When I first started playing SD, I made an SD1.5 version of Lora, but it wasn't released. I used the same materials and a resolution of 1152 to create an SDXL version of Lora. After trying out some large model effects, I found that I needed ADetailer to be more similar. In the next version, I plan to add some new materials and see the effect again.

描述:

训练词语: yaya