中文版本
用于复现画师 Wsman 独特画风的 LoRA 模型。
触发词 (Trigger Word):
wsman style
⚠️ 生成参数特别提醒（重要）
为保证出图质量，强烈建议遵循官方推荐的标准生成配置：
采样器 (Sampler):
Euler迭代步数 (Steps):
40 – 60提示词引导系数 (CFG Scale):
3
ComfyUI 用户注意
许多社区常见的 ComfyUI 模板工作流默认采用了 er_sde 采样方案。该方案与本 LoRA 兼容性极差，会导致包括但不限于以下问题：
严重的人体肢体崩坏
模型生成能力退化
画风特征不明显或无法有效呈现
解决建议： 请务必在采样节点中切回标准 Euler 采样器进行出图。
兼容性说明
Turbo / 蒸馏加速类模型： 暂未经过完整测试，无法保证出图稳定性及风格还原效果。
English Version
A LoRA model designed to faithfully reproduce the signature art style of Wsman.
Trigger Word:
wsman style
⚠️ Crucial Generation Settings (Read Before Generating)
To achieve optimal results, please adhere strictly to the recommended baseline generation parameters:
Sampler:
EulerSteps:
40 – 60CFG Scale:
3
Notice for ComfyUI Users
Many community template workflows in ComfyUI default to the er_sde sampling scheme. This method has very poor compatibility with this LoRA and can cause severe generation artifacts, including but not limited to:
Severe anatomical and limb deformities
General model capability degradation
Muted or failed art style transfer
Recommendation: Switch your sampler node back to standard Euler for consistent, high-fidelity results.
Compatibility Note
Turbo / Fast-inference variants: Untested. Stability and style reproduction on accelerated base models cannot be guaranteed.
Description
相较于SDXL,Anima之类的模型，在如此大的基础模型上训练并不容易。在4090上跑了4个小时左右，经历了多次调参。
使用gemini打标。
建议权重：0.9~0.95
权重过高会出现崩坏。
受限于训练集，模型在皮肤展露较多的情况下表现良好，但是在特殊场景中效果不佳（比如赛博朋克场景）
Translated by Gemini:
Compared to models like SDXL or Anima, training on such a massive base model is no easy task. It took about 4 hours of training on an RTX 4090, going through multiple rounds of parameter tuning.
Captioning: Tagged using Gemini.
Recommended Weight: 0.9–0.95
Setting the weight too high may cause generation artifacts or image breakdown.
Due to training dataset limitations, the model performs well with high skin exposure, but yields subpar results in stylized or specialized settings (such as cyberpunk environments).