Type | |
Stats | 334 86 |
Reviews | (61) |
Published | Oct 24, 2023 |
Base Model | |
Training | Steps: 6 Epochs: 30 |
Usage Tips | Clip Skip: 7 |
Trigger Words | shigure_(\blue_archive\) cyan_hair purple_eyes short_hair bath_yukata yukata low_tied_hair halo |
Hash | AutoV2 2322562ECB |
那么,她的强度怎么样,能打总力战一档吗?
你问我强度,都踏马请你进被窝了你问我她的强度,你***铁南同,你跟一档过一辈子吧,就抱着你那堆破石头,你的学生都是我的了😡
推荐权重为1.0,如果出现过拟合可以适当降低权重
极度推荐使用Lora Block Weight进行分层,推荐预设为1,1,1,1,1,0,0,0,1,1,1,1,1,1,1,0,0
因为没有进行分层而导致的效果不稳定概不负责
主要关键词为shigure_(\blue_archive\), cyan_hair或者aqua_hair, purple_eyes, short_hair, halo, bath_yukata, yukata, striped_yukata,
写上cleavage,bare_shoulders或者off_shoulders之类的tag可以让浴衣脱下一部分
oversized_clothes和loose_yukata也有不错的效果
weasel_tail可能会有效果,但是我没尝试过
光环无法稳定生成,如果可以的话那算你运气好
写上low_tied_hair来加上小辫子
使用anyLoRA为基础进行训练,对不同模型的适配度可能不同,请自行尝试
可能出现人体结构错误,多尝试几次即可,这个我也没法解决
你无法通过本LoRA生成出原版服装,因为训练集里面没有
Recommendation weight is 1.0, if overfitting occurs, the weight can be appropriately reduced.
Highly recommend using Lora Block Weight for layering, recommended preset is 1,1,1,1,1,0,0,0,1,1,1,1,1,1,1,0,0.
We are not responsible for any unstable effects caused by not layering.
The main keywords are 'shigure_(blue_archive)', 'cyan_hair' or 'aqua_hair', 'purple_eyes', 'short_hair', 'halo', 'bath_yukata', 'yukata', 'striped_yukata'
Adding tags like 'cleavage', 'bare_shoulders' or 'off_shoulders' can make part of the yukata come off.
oversized_clothes and loose_yukata also work well.
'weasel_tail' may use, but I haven't tried yet.
The halo cannot be generated stably, so consider yourself lucky if it works.
Adding the tag 'low_tied_hair' can optimize the hairstyle.
Train using anyLoRA as a base, the adaptability may vary for different models, so please try it yourself.
There may be errors in the body structure, when it happens, try again.
You cannot generate the original outfit using this LoRA, as it is not included in the training dataset.
Translated by GPT3.5