# HotNerd-Engineering Male College Student LoRA / 工科男大学生 LoRA
## 中文说明
### 模型简介
本 LoRA 用于生成真实、年轻、帅气的中国工科男大学生形象。人物以普通短黑发、基础校园穿搭、自然体态和清亮眼神为核心；眼镜版采用轻薄镜框，避免呆板感。整体不走偶像造型、成熟职场或夸张健美路线。
### 触发词
engineeringmale
生成时建议把触发词放在提示词最前面。
### 主要控制 TAG
眼镜：
glassesno_glasses
头发：
short_black_hairfringe_black_hairmessy_black_hairspiky_black_hair
体型：
slim_buildlean_buildaverage_buildsturdy_buildathletic_buildbroad_shouldersnarrow_waist
构图和姿势：
full_bodythree_quarter_bodyknee_upfront_viewthree_quarter_viewprofile_viewback_viewstandingwalkingarms_crossedhand_in_pocketholding_backpack_strap
常见服装和物品：
tshirtshirtpolohoodiesweatshirtovershirtjacketpantscargo_pantsjeanssneakersbackpackwatch
服装通常可以加颜色前缀，例如：
blue_gray_tshirtcharcoal_pantsoff_white_sneakersblack_backpack
背景：
studio_background
### 推荐提示词
眼镜全身版：
```text
engineeringmale, glasses, short_black_hair, full_body, three_quarter_view, average_build, broad_shoulders, narrow_waist, blue_gray_tshirt, charcoal_pants, gray_sneakers, black_backpack, studio_background
```
无眼镜行走版：
```text
engineeringmale, no_glasses, fringe_black_hair, full_body, walking, lean_build, green_tshirt, navy_pants, gray_sneakers, black_backpack, studio_background
```
稍结实体型版：
```text
engineeringmale, glasses, short_black_hair, three_quarter_body, sturdy_build, broad_shoulders, narrow_waist, gray_tshirt, navy_pants, studio_background
```
### 使用建议
- 只想生成基础人物时，可以先使用 engineeringmale，再逐步加入控制 TAG。
- glasses 与 no_glasses 不要同时使用。
- 发型和体型标签各选择一种主要类型即可，避免堆叠互相冲突的标签。
- 想要稍厚实但不胖的体型，推荐使用 sturdy_build, broad_shoulders, narrow_waist。
- LoRA 权重建议从 0.7–1.0 开始测试，并根据底模和采样设置调整。
### 训练集信息
- 训练图片：240张
- 配对短标签：240份
- 统一触发词engineeringmale
- 戴眼镜：156张
- 无眼镜：84张
- 12个目录仅用于整理图片，不代表12个不同触发词
- 院系名称和年龄未写入训练标签
- 标签保留眼镜、头发、体型、衣服、姿势和背景等可控特征
## English Description
### Model Overview
This LoRA is designed to generate realistic, young, and handsome Chinese male engineering college students. Its core visual identity includes ordinary short black hair, practical campus clothing, natural body proportions, and bright, attentive eyes. Glasses variants use lightweight frames without creating a dull or overly nerdy expression. The model avoids idol styling, mature office-worker looks, and exaggerated bodybuilding physiques.
### Trigger Word
engineeringmale
Place the trigger word near the beginning of the prompt.
### Main Control Tags
Glasses:
glasses, no_glasses
Hair:
short_black_hair, fringe_black_hair, messy_black_hair, spiky_black_hair
Body type:
slim_build, lean_build, average_build, sturdy_build, athletic_build, broad_shoulders, narrow_waist
Composition and pose:
full_body, three_quarter_body, knee_up, front_view, three_quarter_view, profile_view, back_view, standing, walking, arms_crossed, hand_in_pocket, holding_backpack_strap
Common clothing and accessories:
tshirt, shirt, polo, hoodie, sweatshirt, overshirt, jacket, pants, cargo_pants, jeans, sneakers, backpack, watch
Clothing tags may include a color prefix, for example:
blue_gray_tshirt, charcoal_pants, off_white_sneakers, black_backpack
Background:
studio_background
### Recommended Prompts
Full-body version with glasses:
engineeringmale, glasses, short_black_hair, full_body, three_quarter_view, average_build, broad_shoulders, narrow_waist, blue_gray_tshirt, charcoal_pants, gray_sneakers, black_backpack, studio_background
```
Walking version without glasses:
```text
engineeringmale, no_glasses, fringe_black_hair, full_body, walking, lean_build, green_tshirt, navy_pants, gray_sneakers, black_backpack, studio_background
```
Slightly sturdier build:
```text
engineeringmale, glasses, short_black_hair, three_quarter_body, sturdy_build, broad_shoulders, narrow_waist, gray_tshirt, navy_pants, studio_background
```
Usage Tips
- For a basic result, start with engineeringmale and add control tags gradually.
- Do not use glasses and no_glasses together.
- Select one primary hair type and one primary body type instead of stacking conflicting tags.
- For a slightly fuller but still fit appearance, use sturdy_build, broad_shoulders, narrow_waist.
- A LoRA weight of 0.7–1.0 is a reasonable starting range. Adjust it according to the checkpoint and generation settings.
### Dataset Information
- 240 training images
- 240 paired short-tag captions
- Unified trigger word: engineeringmale
- 156 images with glasses
- 84 images without glasses
- The 12 source folders are organizational categories, not separate trigger words
- Department names and ages are excluded from the captions
- Captions retain controllable features such as glasses, hair, body type, clothing, pose, and background