这是一个面向 18–25 岁中国年轻男性写实肖像的实验性 LoRA。训练集以普通、自然、不过度精修的青年面孔为主，其中约 70% 是县乡或农村生活环境，约 30% 是普通城市环境；18–22 岁约占 72%。
主启动词
chinaregionmale
地域控制词：
-
northeast_china
：东北
-
north_china
：华北
-
east_china
：华东
-
central_china
：华中
-
south_china
：华南
-
southwest_china
：西南
-
northwest_china
：西北
年龄控制词：
-
age_18_22
-
age_23_25
训练标签不区分城乡，也不区分民族。训练集约 70% 是县乡／农村生活环境，因此普通、自然、略带乡土感的气质由总启动词
chinaregionmale
直接学习。建议一次只使用一个地域词。LoRA 权重可先从
0.45–0.75
测试；如果面孔变化不明显再逐步增加，不建议一开始拉得过高。
示例提示词：
chinaregionmale, northeast_china, age_18_22, front, young Chinese man, short black hair, plain T-shirt, village courtyard, natural daylight, realistic documentary portrait
chinaregionmale, south_china, age_18_22, three quarter left, young Chinese man, short black hair, orchard background, natural skin, realistic photo
chinaregionmale, east_china, age_23_25, front, young Chinese man, simple casual clothes, county-town street, soft daylight, realistic portrait
注意：本模型学习的是合成训练图中的创作性地域风格差异，不代表任何地区或民族存在唯一“标准脸”，也不能用于依据照片判断真人的籍贯、民族或身份。不同地区真实人群高度重叠，请把这些标签理解为生图控制词，而不是生物识别结论。
English Description
An experimental realistic-portrait LoRA for Chinese young adult men aged 18–25. About 70% of the dataset uses county-town or rural everyday settings, while about 30% uses ordinary urban settings. Roughly 72% of the subjects are aged 18–22. The intended look is natural, grounded, and lightly documentary rather than polished or influencer-like.
Main trigger:
chinaregionmale
Regional tags:
northeast_china
,
north_china
,
east_china
,
central_china
,
south_china
,
southwest_china
,
northwest_china
.
Age tags:
age_18_22
,
age_23_25
.
The training captions do not separate ethnic groups or rural/urban subjects. Around 70% of the images use county-town or rural settings, so the grounded, slightly rustic everyday look is learned by the main trigger itself. Use one regional tag at a time. A LoRA weight around
0.45–0.75
is a practical starting range.
This model captures creative visual tendencies from synthetic training portraits. It does not define a single “standard face” for any region or ethnic group and must not be used to infer a real person’s origin, ethnicity, or identity from an image. Treat the labels as generation controls, not biometric conclusions.