My pixiv url:（You can see more of my AI artwork here)

https://www.pixiv.net/users/76094070

Update a model trained by LOHA, rather than a v4.0, I prefer to describe it as a Patch, the effect is very average alone, but combined withc v3.0, it will have Nice effect. Here is the combination with v3.0, please adjust the ratio between them according to the preferred effect.

updated 3.0

The use of offset noise and locon, lora better results. Some comparison images↓

I used a game called Destiny XXXXX as training material(100 tachie images), and it looks pretty good.

Sorry about my horrible English.I use google translate.

UPDATED v2.0

This time I did not change the tag word and trained directly. Then I changed some parameters

1. batch size=2

2. network_dim = 128

3. network_alpha = 64

the following is a demonstration picture, in addition to Anything v4.5,

I also tested two other models (dalcefoPainting_2nd and chilloutmix_NiPrunedFp32Fix)You can view the information of my other pictures

↓↓↓↓here is the v2.0_step_effect drawing↓↓↓↓

↓↓↓↓here is the v1.0_step_effect drawing↓↓↓↓

I think that it will make your girls more "oily"

Setting 0.6~0.7 may be a good choice

描述:

训练词语: