Sign In

幻光琉璃 Phantasmal Luminous: The Radiance of Rainbow Dispersion

1.4k
8.9k
137
Type
LoRA
Stats
4,886
Reviews
Published
Oct 11, 2023
Base Model
SD 1.5
Trigger Words
iridescence
Hash
AutoV2
89E0B73F60
default creator card background decoration
XiuAI's Avatar
XiuAI

欢迎返图、一键三连,这对我很重要!!

If there is no problem with your test, please upload a picture, thank you!That's important to me!

If possible, don't forget to order 5 stars⭐️⭐️⭐️⭐️⭐️ and 1 like❤️ ^_^ Please~

需要搬运,商用,请联系我!!

Need transportation, commercial use, please contact me!!

版本信息:

New! v3.0:

由于发现 《奇幻釉-fantastic coloured glaze》配合《幻光琉璃v2.0》可以大大提升服装的丝滑感。我基于2.0+奇幻釉的出图重新训练,并分层融合少量2.0得出了3.0版。

在保留《琉璃》原有特征的前提下,主打写实、丝滑感;修复了前版本身形矮小的问题;减小坏脸坏手的影响,有更好的脸与手;训练集中增加了白色背景;泛用性极强,可搭配各种服装的关键词,得出各式琉璃 欢迎尝试!

On the premise of retaining the original features of 'Liuli', the main focus is on realism and silky smoothness; Fixed the issue of short stature in the previous version; Reduce the impact of bad faces and bad hands, resulting in better faces and hands; Added a white background to the training set; Extremely versatile, can be paired with various clothing keywords, resulting in various types of “liuli” clothes. Welcome to try!

v2.0:

增加了大量服饰版型,能够生成更多样化的裙子

Added a variety of clothing patterns to generate a wider range of dresses

减弱了脸部的影响,使模型更注重服装本身

Reduced the influence on the facial features, making the model focus more on the clothing itself

v1.0:

原型发布

Original release

灵感来源:半透明镭射服,小时候的酸酸糖纸,棱镜效应,光的色散

想象有这么一条裙子也可以有棱镜效应,可以发生光的色散,从而制造出彩虹色系的光感。练了7锅的丹终于最后达到了想要的效果,求一个一键三连!

然而练出模型之后加上机甲类的lora甚至可以出一些水晶炫酷机甲!也期待大家开开脑洞将这个风格与其他任何事物结合,出一些更炫酷的图!

推荐权重:

recommend weight:

v3: 0.7-1.2

v1 or v2: 0.5-0.7

关键提示词:

key prompts:

v3: iridescence

v1 or v2: chromatic dispersion, glowing colors

可选提示词:

Optional prompts:

metallic_lustre, transparent_plastic

调节两者之间的相对权重可以调整不透明的金属质感或透明的塑料质感

Adjusting the relative weight between the two can modify the opaque metallic texture or the transparent plastic feel.

其他值得一试的提示词:

Other prompts worth to try:

coloured glaze, Polychromatic prism effect, rainbowcore, iridescence/opalescence, see_through, aluminum foil, glowing ambiance

不足之处:

V3

背景目前只有黑白二色,对于其他纯色背景适用性不高

V2

这个版本比较注重全身full body的生成(因为训练集多为全身图)。对于脸部特写,半身像的出图质量不高,会出现过拟合。解决方法:可适当减少权重,或者使用 lora block weight 分层。我尝试过的效果不错的两种分层:

OUTS:1,0,0,0,0,0,0,0,0,0,0,0,1,1,1,1,1

CLOTHING:1,1,1,1,1,0,0.2,0,0.8,1,1,0.2,0,0,0,0,0

V1

这个版本lora有些过拟合,会影响脸部,动作,与背景。尝试过分层削弱,但削弱之后反过来影响了服饰的风格效果。所以我就直接上传了不削弱的版本,因为这样的服饰风格我认为才到位。

如果介意影响脸部,可以尝试局部重绘换脸

特别鸣谢:

试试他的模型!给他们点个赞!