Chinese Male Short Hairstyles｜中国男生短发合集

建议：建议搭配其他面容LoRA使用。
PS：Do use with other Face LoRA.

中文说明：

这是一个面向中国年轻男性的短发控制 LoRA，训练集包含约 260 张合成头像，共 13 种发型，每种约 20 张。训练图覆盖正面、三分之二侧面、左右侧面和后侧视角，重点学习发型轮廓、刘海、分缝、顶部纹理以及两侧渐变，而不是某一张固定面孔。

整体风格以 18–25 岁中国男性为主，偏清爽、自然、运动和阳刚，不以长发、夸张烫发或网红造型为主要方向。

基础触发词：

chinesemalehair

发型标签：

• buzz_fade：渐变短寸

• crew_cut：清爽短寸

• short_textured_hair：短纹理发

• short_crop：短裁剪发型

• front_spike_hair：前刺发型

• subtle_part_spike：微侧分前刺

• short_three_seven：短三七分

• short_side_swept：短侧扫发型

• short_weifen_suigai：短微分碎盖

• short_textured_fringe：短纹理刘海

• low_taper_textured_crop：低渐变纹理短裁

• short_six_four_part：短四六分

• straight_short_front_lift：直发短前抬

推荐写法：

chinesemalehair, short_weifen_suigai, young adult Chinese man, natural short black hair, headshot portrait

通常使用“总触发词 + 一个发型标签”即可。多个发型标签可以尝试混合，但可能导致发型结构不稳定。建议先从 LoRA 权重 0.7–0.9 测试，再根据所用底模调整。

训练时对眼睛、鼻子、嘴唇、耳朵和整体气质进行了分别标注，目的是尽量减少发型与固定脸型的绑定。如果人物面貌变化过大，可以降低 LoRA 权重，并在提示词中明确指定需要的脸型和五官。

这是早期版本，不同底模对发型标签的响应可能有所区别，欢迎分享测试结果和改进建议。

English Description:

This is a short-hairstyle control LoRA designed primarily for young Chinese men. The dataset contains approximately 260 synthetic portraits across 13 hairstyles, with around 20 images per hairstyle.

The training images cover front, three-quarter, profile, and rear-side views. The model focuses on hairstyle silhouette, fringe shape, parting, top texture, and side tapering instead of memorizing one fixed identity.

The overall direction is clean, natural, sporty, and masculine, mainly intended for Chinese men aged 18–25. Long hair, exaggerated perms, and overly polished influencer hairstyles are not the primary focus.

Base trigger:

chinesemalehair

Hairstyle tags:

• buzz_fade

• crew_cut

• short_textured_hair

• short_crop

• front_spike_hair

• subtle_part_spike

• short_three_seven

• short_side_swept

• short_weifen_suigai

• short_textured_fringe

• low_taper_textured_crop

• short_six_four_part

• straight_short_front_lift

Example prompt:

chinesemalehair, short_weifen_suigai, young adult Chinese man, natural short black hair, headshot portrait

Using the base trigger together with one hairstyle tag is recommended. Mixing multiple hairstyle tags is possible, but may produce less predictable hair structures. Start with a LoRA weight around 0.7–0.9 and adjust according to the checkpoint.

Eyes, nose, lips, ears, and overall facial impression were captioned separately during dataset preparation to reduce unwanted binding between a hairstyle and a fixed face. If the LoRA changes the character’s identity too strongly, lower its weight and describe the desired facial features more explicitly.

This is an early release. Results may vary across different checkpoints, and feedback is always welcome.