Sign In

szcb911 / chen bin style - 鬼针草风格 for MOSS

Updated: Oct 22, 2026

stylechen binszcb911

Download

1 variant available

bf16 SafeTensor

szcb911_lora_r64.safetensors

BF16, good balance • 286.15 MB

Verified:

Type
LoRA
Stats

40

Reviews
Published

Oct 22, 2026

Base Model

MOSS Image

Training
Steps: 6,000
Usage Tips
Strength: 1
Hash
AutoV2
C984A44D9E
Trigger Words
@szcb911style
_season_spring_159223854.png

⭐If you have the means, please do support the original artist's creative work!
⭐如果您有条件,请支持原创艺术家的创作!

A LoRA that reproduces the drawing style of the artist 鬼针草 (chen bin / szcb911).
一个目的是再现鬼针草(chen bin / szcb911)绘画风格的LoRA。

Artist page: https://www.pixiv.net/users/6049901

I'm re-uploading because there were some "little" issues

重新上传是因为一些 “小” 问题

Trigger word: @szcb911style — put it at the very beginning of your prompt

触发词:@szcb911style — 请将其置于你的提示词开头

V2 CHANGELOG: Using KohakuMoss as the training tool, we readjusted the training parameters and training dimensions. v1 employed a training resolution of 1m , while v2 directly used an almost full-image resolution of 2m for training. v1RANK32→v2RANK64. Use the DoRa format without adding new training layers. This approach enhanced the accuracy of composition while ensuring compatibility.The characteristic sole color in chen bin's works may slightly differ from that of V1.

V1 TIPS: the stability of the foot in the generated image from the model is slightly unsatisfactory. I need to try several more seeds, The color is slightly different from v2

V2变更内容:换用KohakuMoss作为训练工具,重新调整了训练参数与训练尺寸
v1是1m训练分辨率,v2则直接是2m,几乎用原图分辨率进行训练,v1的RANK32→v2的RANK64,并使用了DoRa格式,但没有增加新的训练层,在提升肢体的稳定性的同时确保了兼容性。鬼针草老师标志性的脚底颜色会与V1略微有些差别。

v1提示:模型的肢体稳定性稍微有些问题,需要多抽几次seed。


Comparison Chart/ 对比图 :
test prompt / 测试提示词:1girl, solo, pantyhose, white pantyhose, feet, soles, toes, sitting, on floor, 她坐在浅色的地板上,双腿向前伸直,双脚脚底正对镜头,十个脚趾微微张开,极薄的白丝紧贴脚底,脚趾和脚心肤色从丝袜下透出来,每根脚趾之间能看到丝袜收紧的细褶,她半眯着眼微微张开嘴轻笑,面带红晕,双手撑在身后,身体微微后仰,侧光从斜后方勾出腿部与脚底的轮廓,背景虚化成柔和的浅色。
The actual measurement shows that v2 vs v1 has reduced the occurrence of bad toes/extra toes.

实测v2 vs v1一定程度减小了脚趾崩坏的情况。

Base model: MOSS image base

Training data

200 images from the artist's posts with perform manual full-scale cleaning on the tags

基础模型:MOSS image base

训练数据

200张图片进行手动的全量tag清洗