中文介绍这是一个面向 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<code>chinaregionmale</code>地域控制词：- <code>northeast_china</code>：东北- <code>north_china</code>：华北- <code>east_china</code>：华东- <code>central_china</code>：华中- <code>south_china</code>：华南- <code>southwest_china</code>：西南- <code>northwest_china</code>：西北年龄控制词：- <code>age_18_22</code>- <code>age_23_25</code>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 <code>chinaregionmale</code> 直接学习。建议一次只使用一个地域词。LoRA 权重可先从 <code>0.45–0.75</code> 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：<code>chinaregionmale, northeast_china, age_18_22, front, young Chinese man, short black hair, plain T-shirt, village courtyard, natural daylight, realistic documentary portrait</code><code>chinaregionmale, south_china, age_18_22, three quarter left, young Chinese man, short black hair, orchard background, natural skin, realistic photo</code><code>chinaregionmale, east_china, age_23_25, front, young Chinese man, simple casual clothes, county-town street, soft daylight, realistic portrait</code>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 DescriptionAn 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: <code>chinaregionmale</code>Regional tags: <code>northeast_china</code>, <code>north_china</code>, <code>east_china</code>, <code>central_china</code>, <code>south_china</code>, <code>southwest_china</code>, <code>northwest_china</code>.Age tags: <code>age_18_22</code>, <code>age_23_25</code>.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 <code>0.45–0.75</code> 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.

Files with SHA256: 097f4d1c422d4404281b0ca248dc15406de3272b9ccd04f60ac6153b63538765