Kiritani Haruka _ Kiritani Haruka _ Kirita Haruka

Kiritani Haruka

has a total of five trigger words Kiritani Haruka, which can generate all kinds of Kiritani Haruka. Simple tags would have a better effect little

by little

Faces

, that is, only faces, can be used to generate emoticons or the like. To generate images of other parts, you need to use a large number of tags. There are many things that can be generated, and idol clothes should also be better. This is a singing suit. In fact, there are many pictures of this suit, but the quality is uneven. If the picture runs out far from the original

one, I recommend

Animated tag: skirt, bow, sleeveless, frilled skirt, idol, shirt, hair bow, white footwear, frills, white shoes, vest, layered skirt, hair ornament, clover serafuku, this

It's a school uniform. I've found relatively few pictures of this set of clothes. The main problem with the running map is that the color of the clothes sometimes doesn't require manual tagging: grey skirt, grey shirt swimsuit: There are even fewer Kiritani Haruka on this original swimsuit, but the image quality is pretty good. Barely enough, it's a bit restorative. If you want to run out of the other swimsuits, Haruka Kiritani uses Kiritani Haruka, and Face

Add the swimsuit tag you want, the

recommended Lora weight is 0.7-0.8. If I use 0.7 more to run real images, my hair is often black, so I need to add my own tag: blue hair, and the tag weight

is 1.4

The sample pictures are five sets of clothes for each of the eight big models. They are all 1080P pictures. The next thing I need

to do is a PJSK painting style Lora and Big Black Fox 2023.6.22

The first version of the new Run is almost the same as the previous one. The main thing is that the accuracy of the uniform and song suit is improved, and it can be generated relatively well by just writing the trigger word, but since there are fewer pictures of the swimsuit and the number of steps, it is overfitting, and then there are times when running the real model is prone to problems, so the real model will be more stable using the previous version. It's OK if the two versions are different, OK?