ghost
Sign In

Seoyeon Yoo | 서연 유 FROM Succubus System | StyleIncluded |

66
294
108
6
Verified:
SafeTensor
Type
LoRA
Stats
294
108
Reviews
Published
Nov 7, 2023
Base Model
SD 1.5
Usage Tips
Clip Skip: 2
Strength: 0.8
Trigger Words
SeoyeonYoo
SucSysStyle
Hash
AutoV2
414402DDDB
default creator card background decoration
First Birthday Badge
kurotsune's Avatar
kurotsune

This model is the 2nd one of the Succubus System serie. It is aimed to generate images of Seoyeon Yoo in a style that somewhat resemble the manhwa.

Compared to the minah kim one, it's pretty versatile (even more in NSFW stuff), but it's very prompt dependant.

Due to the high number of training images used, it has some weight issues, that need to be tweeked to make it work properly.

The Character keyword is: "SeoyeonYoo", it works perfectly in lora strenght between 0.5 and 1.

The style keyword is the same of the Minah Kim one: "SucSysStyle", and it works decently (You might want to set this one at around 1.3- 1.4 when the lora strenght is lower).

That's it, The model is pretty much plug&play, it doesn't require particoular tags, but it works better with nsfw ones.

Settings:

Lora strenght:                  |  0.5 -  1  |
SeoyeonYoo tag strenght:        |      1     |
SucSysStyle tag strenght:       |  0.8 - 1.4 |

Tips and tricks:

Here are some tags that might be usefull (but not needed):

black hair, purple eyes, blush, mature female, 1girl, bangs, large breasts

Here are some suggested negative tags:

text, korean text, speech bubble, closed eyes (it is suggested if you want to have an image with better open eyes, because a lot of the training images had closed eyes tag)

[Disclaimer]

The serie Succubus System has a really big flaw when it comes to images for training, those are generally close-up, or even worse the faces and other items get often deformed. Those images were needed to get the style, but they inevitably might affect the output of other things.

So be warned about the fact that you might need few tries before reaching the exact output you wanted (generally speaking the Character works fine, but the composition might get affected).

This model will eventually get updated, when more images/more training data will be avaible.

Hope you like it (even toh it could be better).