Krea2 Opaque Pantyhose 厚裤袜
详情
下载文件 (1)
关于此版本
模型描述
V1 — Krea2-pantyhose-V3_B-000004(当前发布版)
训练日期: 2026-07-12 数据集: 116 张,4 色 × 2 光泽,不透明 打标: 自然语言英文句子 训练框架: Fizgig(Krea 2 RAW) 推荐检查点: 第 4 轮(loss 0.0656) 备用检查点: 第 3 轮(loss 0.0689)
黑色效果最稳,高权重下依然保持厚实质感
白/灰/肉色高权重时也能出厚裤袜效果,但不如黑色稳定
权重范围 0~3,越高越厚实不透肉
底模必须是 Krea 2 Turbo,8 步推理
后续计划:如果大家喜欢的话可以给我点赞或者留言,我会尝试去优化这个lora
V1 — Krea2-pantyhose-V3_B-000004 (Current Release)
Training date: 2026-07-12 Dataset: 116 images, 4 colors × 2 finishes, opaque only Captions: Natural language English sentences Training framework: Fizgig (Krea 2 RAW) Recommended checkpoint: Epoch 4 (loss 0.0656) Alternative checkpoint: Epoch 3 (loss 0.0689)
Black performs best — stays thick and solid even at high weights
White/grey/nude can also achieve thick opaque results at higher weights, but less consistently than black
Weight range: 0–3, higher = thicker and more opaque
Base model must be Krea 2 Turbo, 8-step inference
Trigger word:
pantyhoseFuture plans: If you like this LoRA, feel free to leave a like or comment. I'll keep improving it based on feedback.
背景
在 C 站寻找 Krea 2 的裤袜 LoRA 时,我发现已有的裤袜 LoRA 虽然效果不错,但在厚裤袜(不透明)的表现上稍微没有那么给力。所以开始自己学习如何训练 LoRA、如何选择训练集以及如何打标。这个 LoRA 是我第一个发布的 LoRA,可能有很多不足之处,也希望各位能给我指出和建议。
功能介绍
不透明材质 — 厚实不透肉,颜色饱满均匀
哑光 vs 油光 — 两种光泽效果清晰可辨
4 种基础颜色:黑色、白色、灰色、肉色
细节丰富 — 脚底防滑条、面料褶皱、紧身包裹感一应俱全
使用方式
底模: Krea 2 Turbo(训练基于 Krea 2 RAW)
触发词:
pantyhose推荐权重: 0–3.0(权重越高裤袜越厚实不透肉)
步数: 8 步(Turbo)
注意: 黑色效果最佳,高权重下依然保持厚实哑光/油光质感。其他颜色(白/灰/肉色)在高权重时也能达到厚裤袜效果,但不如黑色稳定。
附加效果
除了控制厚度和光泽,这款 LoRA 还能改善裤袜的纹理细节——包括面料光泽感、编织纹理和腿部缝合线的真实感,让裤袜看起来更有材质质感而非简单的皮肤上色。
提示词参考 — 8 种组合
组合关键词黑色哑光不透明black matte opaque黑色油光不透明black glossy opaque灰色哑光不透明grey matte opaque灰色油光不透明grey glossy opaque白色哑光不透明white matte opaque白色油光不透明white glossy opaque肉色哑光不透明nude matte opaque肉色油光不透明nude glossy opaque
训练参数
数据集:116 张,4 色 × 2 光泽
打标方式:自然语言句子
分辨率:1024×1024
训练框架:Fizgig(Krea 2 RAW)
Keea2 Opaque Pantyhose LoRA
A Krea 2 LoRA trained exclusively on opaque pantyhose in 4 colors × 2 finishes, giving you precise control over hosiery in your generations.
Backstory
When I searched for Krea 2 pantyhose LoRAs on Civitai, I found some great options, but most of them didn't perform as well on thick/opaque pantyhose specifically. So I started learning how to train LoRAs — how to curate datasets, how to caption them properly, and how to tune the training process. This is my very first published LoRA, and there's definitely room for improvement. Any feedback and suggestions are welcome!
Features
Opaque only — Solid color coverage, non-transparent
Matte vs Glossy — Two finishes clearly distinguishable
4 base colors: black, white, grey, nude
Rich details — Grip soles, fabric folds, skin-tight fit
How to Use
Base model: Krea 2 Turbo (trained on Krea 2 RAW)
Trigger word:
pantyhoseWeight range: 0–3.0 (higher = thicker, more opaque)
Steps: 8 (Turbo)
Note: Black works best at high weights. Other colors (white/grey/nude) can achieve thick opaque results at higher weights but less consistently than black.
Additional Effects
Beyond opacity and gloss control, this LoRA improves the texture details of pantyhose — fabric sheen, knit texture, and seam lines — making hosiery look like real fabric rather than painted-on skin.
Prompt Reference — 8 Combinations
LookKeywordsBlack matte opaqueblack matte opaqueBlack glossy opaqueblack glossy opaqueGrey matte opaquegrey matte opaqueGrey glossy opaquegrey glossy opaqueWhite matte opaquewhite matte opaqueWhite glossy opaquewhite glossy opaqueNude matte opaquenude matte opaqueNude glossy opaquenude glossy opaque
Training Details
Dataset: 116 images, 4 colors × 2 finishes
Captions: Natural language sentences
Resolution: 1024×1024
Framework: Fizgig (Krea 2 RAW)








