以下是针对Stable Diffusion 3文生图模型的人物写真Prompt的设计方法论，结合摄影原理与AI生成特性，精准捕捉人物美感：
一、核心美学要素构建
- 黄金比例构图
• 采用rule of thirds构图法：portrait of a woman:1.2, (three-quarters profile:0.9), (face centered in upper third:1.1)• 动态平衡示例：sitting on wooden bench, (left hand touching chin:0.7), (right leg crossed:0.6), (flowing hair:1.3) - 光影美学公式
• 伦勃朗光：Rembrandt lighting, (soft rim light:0.8), (gentle shadow on collarbone:1.2)• 电影级布光：cinematic three-point lighting, (volumetric dust particles:0.7)
二、人物特征强化策略
3. 五官特写模块
(almond-shaped eyes:1.4) with (catchlights:0.9),
(defined jawline:1.2) contrasting with (soft cheekbones:0.8),
(smile showing dimples:0.7),
(glossy lips:1.3) with (lipstick sheen:0.6)- 皮肤质感处理
• 写实皮肤：hyper-detailed skin texture, (subsurface scattering:1.1), (pore-level details:0.9)• 艺术化处理：oil painting skin texture with visible brushstrokes, (velvet skin:0.7)
三、风格化表达技巧
- 跨媒介融合
• 摄影级写实：8k Hasselblad X2D sample, (film grain:0.6), (dynamic range:1.3)
• 插画美学：watercolor paper texture overlay, (cel-shading:0.8), (ink wash details:0.7)
- 情绪氛围营造
• 温柔气质：soft focus effect, (pastel color grading:1.2), (gentle bokeh:0.9)• 神秘气质：dramatic chiaroscuro lighting, (smoke haze:0.7), (vintage film filter:0.6)
四、SD3专属优化方案
- 多模态控制
ControlNet提示：
full-body skeleton pose (sitting:1.2)
主提示：
(exotic dancer:1.3) in motion capture suit,
(intricate feather costume:0.9),
(motion blur:0.7) on (neon-lit stage:1.1),
unreal engine 5 nanite details- 动态美学参数
• 渐变表情：[smile intensity:0.3→0.8]• 时空转换：time-lapse transition from dawn to dusk:0.6
五、实战案例模板
日系清新人像(japanese idol:1.4), (soft pastel background:0.9), (spring cherry blossom:0.7), (wind-blown hair:1.2), (delicate freckles:0.8), (film simulation:1.3), (exposure:0.6), (sharp focus on eyes:1.1)
复古胶片质感(1950s pin-up style:1.3), (vintage red lipstick:0.9), (curler waves:1.2), (silk robe texture:0.8), (kodak portra 400 film:1.1), (70s studio lighting:0.7)
六、避坑指南
- 权重平衡公式
(主体特征:1.2) + (环境特征:0.8) + (风格修饰:0.6) = 2.6（总权重建议不超过3.0） - 负面提示组合
nsfw, deformed fingers, (worst quality:1.4), (blurry face:1.3), (extra limbs:1.2), (text errors) - 迭代参数建议
• 基础步数：25-30步（人像细节收敛） • 采样器：DPM++ 2M Karras（平衡速度与质量） • 分辨率：768x1152（竖版人像最佳比例）
建议配合ComfyUI工作流进行模块化测试，通过节点可视化调整各要素权重。最新SD3对中文提示词理解提升显著，可尝试混合中英文描述获得更精准结果（例：（水墨晕染效果:1.2）ink diffusion effect）。