Type | |
Stats | 1,057 5,857 |
Reviews | (171) |
Published | May 10, 2023 |
Base Model | |
Training | Steps: 2,060 Epochs: 5 |
Trigger Words | 1girl multiple girls |
Hash | AutoV2 A0C4D72D0E |
训练集图片来自https://www.pixiv.net/artworks/108131440
Training set images from https://www.pixiv.net/artworks/108131440
推荐使用deliberate_v2模型,它完美还原了我训练集中的人物脸部特征,这根那些卡通模型有很大的区别。你可以参考我最近上传的样品图片,它在下面,只有几张。
deliberate_v2 model is recommended, it perfectly reproduced the face features in my training set, which is much different from the cartoon model. You can refer to the sample picture I uploaded recently, it is below, there are only a few.
我在960x512训练,你也可以试试别的
只需要1girl,或multiple girls就能得到这样一个场景
我的训练集风格是有些不同的,但是我产出的图片风格是一致的,我觉得没有训练下去的必要了。
它的手有些糟糕
I do 960x512 training. You can try something else
It only takes 1girl, or multiple girls, to get a scene like this
My training set style is a little different, but the style of the pictures I produce is consistent, and I don't see the need for further training.
It has some bad hands
我测试了一些我训练集里面有特征的图片,得到了一些能够控制动作的可选tag:
I tested some of the featured images in my training set and got some optional tags to control the action:
===multiple girls=============
side by side,
back to back,
jumping
=========1girls==============
mushroom,
running,
jumping,
from side,look back,
waving,
reaching towards viewer, ======》 (It works, but it's not good)
hand on own face,
grin,
on hand on own face,standing,look back,
light smile