Type | |
Stats | 913 |
Reviews | (101) |
Published | Feb 24, 2023 |
Base Model | |
Training | Steps: 6,400 Epochs: 20 |
Trigger Words | 1girl broom shirt sweater witch_hat |
Hash | AutoV2 2CFABC8B1D |
elaina_v3已推出,你依然可以使用elaina_v2,V2.0版本对画风可能更为还原。
elaina_ v3 has been launched, you can still use elaina_ v2. The V2.0 version may restore the painting style more.
V3.0:https://civitai.com/models/15424/majo-no-tabitabi-elaina-v30
V2.0 中文介绍
本Lora进行了6400步的训练以及逐一tag调试,底模为Anything-v4.5,本Lora意在还原小说画风,使用它你并不能生成动画画风与其他画风。
你可以根据下列提示词改变伊蕾娜的衣服风格或者增加扫帚与魔女帽:
魔女服 shirt、毛衣 sweater、扫帚 broom、魔女帽 witch_hat
要生成伊蕾娜你至少需要的提示词是:
1girl、elaina
扫帚经常会一分为二,以及关于魔女胸针我并没有指定相应的tag,可能会有不佳的表现,你可以通过加大权重后局部重绘修复胸针,以后的版本我会尽可能改进。
头发颜色与长度、眼睛颜色是固有属性,你不能很好的通过提示词改变。
本Lora使用的大概权重在0.85-0.95,当然根据不同的采样方法和提示词相关性也会出现其他可行的数值。
V2.0 English introduction
This Lora has carried out 6400 steps of training and tag-by-tag debugging, and the basic model is Anything-V4.5, this Lora is intended to restore the style of novel painting. You can't generate animation style and other styles with it.
You can change Elaina's dress style or add brooms and witch hats according to the following prompts:
shirt,sweater,broom,witch_hat
To generate Irina, at least the prompt words you need are:
1girl,elaina
The broom is often divided into two parts, and I have not specified the corresponding tag for the witch brooch, which may have poor performance. You can repair the brooch by increasing the weight and inpaint. I will try to improve it in the future version.
Hair color and length as well as eye color are inherent attributes, and you can't change them well through prompt words.
The approximate weight used in this Lora is 0.85-0.95. Of course, other feasible values will also appear according to different Sampling steps and CFG Scale.