[主体 Subject]
  ↓
[动作/状态 Action]
  ↓
[环境/场景 Environment]
  ↓
[视觉层级与构图 Composition]
  ↓
[风格/艺术方向 Style]
  ↓
[摄影/渲染参数 Technical Details]
  ↓
[质量与限制条件 Quality Constraints]

[主体] + [动作/场景] + [构图/镜头] + [光影/色彩] + [艺术风格] + [渲染参数]

A cyberpunk city

A futuristic cyberpunk city (主体+环境),
a lone female hacker standing on a rooftop (主体+动作),
foreground focus, background city blurred (视觉层级),
neon lights, rainy night, reflections (风格),
cinematic lighting, shallow depth of field, 35mm lens (摄影术语),
ultra-detailed, 8k, realistic rendering (质量控制)

错误顺序：
"35mm lens, neon lights, a hacker, cyberpunk city"

正确顺序：
"A cyberpunk city (全局氛围)
→ with neon lights and rain (环境细节)
→ a hacker on rooftop (主体)
→ 35mm lens, cinematic (技术参数)"

interface PromptStructure {
  subject: string;        // P0 - 必需
  action?: string;        // P1 - 推荐
  environment: string;    // P0 - 必需
  composition: string;    // P1 - 推荐
  style: string;          // P0 - 必需
  technical: string;      // P2 - 可选
  quality: string;        // P1 - 推荐
}

const prompt: PromptStructure = {
  subject: 'A lone astronaut',
  action: 'floating in space',
  environment: 'surrounded by distant galaxies',
  composition: 'centered, rule of thirds',
  style: 'sci-fi concept art, dramatic',
  technical: '85mm lens, f/2.8, rim lighting',
  quality: 'ultra-detailed, 8k, photorealistic'
};

// 组装成最终 Prompt
const finalPrompt = Object.values(prompt).filter(Boolean).join(', ');

❌ 错误示例：
"beautiful amazing stunning gorgeous epic masterpiece 
ultra-detailed 8k photorealistic cinematic dramatic..."

✅ 正确示例：
"Photorealistic portrait, cinematic lighting, 
ultra-detailed, 8k resolution"

Foreground (前景)
  ↓
Midground (中景)
  ↓
Background (背景)

"A medieval knight in the foreground (sharp focus),
a castle in the midground (slightly blurred),
mountains in the background (soft focus, atmospheric perspective)"

(main subject:1.5), sharp focus,
(background:0.8), blurred,
bokeh effect

main subject, sharp focus --no blurred background

"Portrait of a woman,
rim light from the left (highlighting hair),
key light from front-right (soft, diffused),
backlight creating subtle glow,
golden hour atmosphere,
cinematic lighting"

Close-up (特写)
  ↓
Medium Shot (中景)
  ↓
Wide Shot (远景)
  ↓
Extreme Wide Shot (大远景)

"Close-up portrait of an elderly man,
wrinkles and details visible,
shallow depth of field,
85mm lens, f/1.8"

Large Aperture (f/1.4 - f/2.8)
→ Shallow depth of field (浅景深)
→ 主体清晰，背景虚化
→ 适合人像、产品

Small Aperture (f/8 - f/16)
→ Deep depth of field (深景深)
→ 全局清晰
→ 适合风景、建筑

"Product photography of a luxury watch,
macro lens, f/2.8,
soft lighting from top-left,
shallow depth of field,
reflections on polished surface,
studio setup, white background,
ultra-detailed, 8k"

Classical Styles:
- Renaissance
- Baroque
- Impressionism
- Art Nouveau

Modern Styles:
- Minimalism
- Abstract
- Surrealism
- Pop Art

Contemporary:
- Cyberpunk
- Steampunk
- Vaporwave
- Low Poly

Lighting:
- Global Illumination (全局光照)
- Ambient Occlusion (环境光遮蔽)
- Subsurface Scattering (次表面散射)
- Volumetric Lighting (体积光)

Materials:
- Physically Based Rendering (PBR)
- Metallic (金属)
- Dielectric (电介质)
- Translucent (半透明)

"3D render of a futuristic car,
octane render,
physically based materials,
metallic paint with clear coat,
global illumination,
volumetric fog,
studio lighting setup,
ultra-realistic, 8k"

✅ 概念设计
✅ 插画、封面、艺术创作
✅ 情绪化、氛围感强的作品
✅ 需要高美学质量的场景

/imagine prompt: [description] --ar 16:9 --stylize 500 --v 6

/imagine prompt: iridescent dragon scales, 
brutalist architecture background, 
ethereal glow, volumetric fog, 
cinematic composition, 
ultra-detailed textures 
--ar 16:9 --stylize 750 --v 6 --no people, text

✅ 产品图
✅ 信息图、教育内容
✅ UI/UX 设计素材
✅ Marketing 视觉
✅ 需要精确语义的场景

"A modern minimalist living room with large windows,
natural sunlight streaming in,
Scandinavian furniture in light wood and white fabric,
indoor plants on wooden shelves,
clean lines and uncluttered space,
professional interior photography"

✅ 定制风格模型
✅ 批量生成
✅ 需要严格遵循设计规范的生产环境
✅ 研究与二次开发
✅ 像素级构图控制

Positive Prompt: [想要的内容]
Negative Prompt: [不想要的内容]

(keyword:1.5)  # 增强权重
(keyword:0.8)  # 降低权重
[keyword]      # 降低权重（简写）

Positive Prompt:
"(masterpiece:1.4), (best quality:1.4),
portrait of a young woman,
(detailed face:1.3), (expressive eyes:1.2),
natural lighting, soft shadows,
85mm lens, f/1.8, shallow depth of field,
photorealistic, ultra-detailed"

Negative Prompt:
"(worst quality:1.4), (low quality:1.4),
blurry, distorted, deformed,
bad anatomy, extra fingers,
watermark, signature, text"

Parameters:
Steps: 30
CFG Scale: 7.5
Sampler: DPM++ 2M Karras
Seed: 12345678

Midjourney:
"ethereal forest, bioluminescent plants, 
mystical atmosphere --ar 16:9 --stylize 750"

DALL·E 转换:
"A mystical forest scene with glowing bioluminescent plants,
ethereal atmosphere with soft blue and green lights,
magical and dreamlike quality,
wide-angle view, cinematic composition"

DALL·E:
"A cozy coffee shop interior with warm lighting,
wooden furniture, and people reading books"

Stable Diffusion 转换:
Positive:
"(cozy coffee shop interior:1.3),
(warm ambient lighting:1.2),
wooden tables and chairs,
people reading books,
soft focus background,
professional photography,
(detailed textures:1.1)"

Negative:
"blurry, low quality, distorted,
bad anatomy, watermark"

Subject + Photography Terms + Lighting + Lens + Quality

"Photorealistic [subject],
[lens specification],
[lighting description],
[depth of field],
[additional details],
ultra-detailed, 8k, professional photography"

"Photorealistic portrait of a 30-year-old woman,
85mm lens, f/1.8,
natural window light from the left,
shallow depth of field, bokeh background,
detailed skin texture, expressive eyes,
professional headshot,
ultra-detailed, 8k"

"Product photography of a luxury smartphone,
macro lens, f/2.8,
studio lighting with softbox,
white background, reflections on glass surface,
detailed metal frame, screen reflection,
commercial photography,
ultra-detailed, 8k"

"Architectural photography of modern glass building,
wide-angle lens, 16mm,
golden hour lighting,
deep depth of field, sharp throughout,
dramatic sky, reflections on glass,
professional architectural photography,
ultra-detailed, 8k"

Subject + Art Medium + Color Palette + Line Style + Mood

"[art medium] illustration of [subject],
[color description],
[line style],
[mood/atmosphere],
[additional artistic elements]"

"Watercolor illustration of a peaceful garden,
soft pastel colors, gentle brush strokes,
transparent washes, flowing textures,
dreamy and ethereal atmosphere,
children's book style"

"Flat design illustration of a city skyline,
bold colors, geometric shapes,
clean lines, no gradients,
modern and minimalist,
vector art style"

"Hand-drawn ink illustration of a fantasy dragon,
detailed line work, cross-hatching,
black and white, traditional pen and ink,
intricate scales and textures,
classic illustration style"

Subject + Rendering Engine + Materials + Lighting + Technical Quality

"3D render of [subject],
[rendering engine],
[material description],
[lighting setup],
[technical specifications],
ultra-realistic, 8k"

"3D render of a futuristic sports car,
octane render,
metallic paint with clear coat,
carbon fiber details,
studio lighting with HDRI,
physically based materials,
global illumination,
ultra-realistic, 8k"

"3D architectural render of modern villa,
V-Ray render,
realistic materials (concrete, glass, wood),
natural daylight simulation,
ambient occlusion,
photorealistic vegetation,
ultra-detailed, 8k"

"3D character render of a sci-fi soldier,
Unreal Engine 5,
PBR textures, subsurface scattering,
dramatic rim lighting,
detailed armor and weapons,
game-ready quality,
4k textures"

"Cyberpunk cityscape (主风格),
watercolor painting style (艺术媒介),
soft pastel colors (色彩调整),
dreamy atmosphere (情绪)"

"Viking warrior (历史元素),
in Pixar animation style (现代风格),
3D render, octane (技术实现),
vibrant colors, playful (情绪调整)"

"Japanese temple architecture (东方元素),
Art Nouveau decorative style (西方艺术),
intricate patterns, flowing lines,
harmonious fusion"

"Steampunk airship (蒸汽朋克),
rendered in Studio Ghibli animation style (吉卜力风格),
soft watercolor textures,
warm sunset lighting,
whimsical and nostalgic atmosphere,
detailed mechanical parts with organic flowing design,
hand-painted quality"

❌ 错误思维:
"希望模型能理解我想要的感觉"

✅ 正确思维:
"用明确的限制条件定义输出"

❌ 抽象描述:
"Beautiful landscape"

✅ 具体描述:
"Mountain landscape at golden hour,
snow-capped peaks, alpine meadow with wildflowers,
dramatic clouds, warm sunlight,
wide-angle view, 24mm lens"

❌ 错误方式:
"A forest without people"

✅ 正确方式:
Positive: "A pristine forest, untouched wilderness"
Negative (SD): "people, humans, figures"

1. 基础 Prompt
   ↓
2. 生成初版
   ↓
3. 分析问题（构图/光线/细节）
   ↓
4. 调整 Prompt
   ↓
5. 固定 Seed，微调参数
   ↓
6. 重复直到满意

// 固定随机性，实现可复现
const config = {
  prompt: "...",
  seed: 12345678,  // 固定种子
  steps: 30,
  cfg: 7.5
};

// 相同配置 → 相同输出
// 微调 Prompt → 可控变化

优化前:
"A dragon"

优化后:
"Majestic dragon with iridescent scales,
perched on ancient ruins,
dramatic storm clouds,
epic fantasy atmosphere,
cinematic composition
--ar 16:9 --stylize 750"

优化前:
"Cool tech gadget"

优化后:
"A sleek wireless charging pad,
minimalist design with matte black finish,
LED indicator light,
placed on a modern desk,
product photography with soft lighting"

优化前:
"Beautiful woman"

优化后:
Positive:
"(masterpiece:1.4), (best quality:1.4),
portrait of elegant woman,
(detailed face:1.3), (expressive eyes:1.2),
natural makeup, soft smile,
professional photography,
85mm lens, f/1.8,
(photorealistic:1.3)"

Negative:
"(worst quality:1.4), (low quality:1.4),
blurry, distorted, deformed,
bad anatomy, extra limbs,
watermark, text"

prompt_record:
  id: "PR-2024-001"
  date: "2024-12-26"
  model: "Midjourney V6"
  
  prompt:
    positive: "..."
    negative: "..."
    parameters:
      ar: "16:9"
      stylize: 750
      quality: 1
  
  result:
    image_url: "..."
    rating: 8/10
    notes: "Good composition, needs more detail in background"
  
  iterations:
    - version: 1
      changes: "Added 'cinematic lighting'"
      result: "Improved mood"
    - version: 2
      changes: "Increased --stylize to 750"
      result: "Better artistic quality"

v1.0: 基础 Prompt
v1.1: + 光线描述
v1.2: + 构图控制
v1.3: + 质量参数
v2.0: 重构结构，优化顺序

Compatible:
✅ Cyberpunk + Neon + Rain
✅ Fantasy + Epic + Dramatic
✅ Minimalist + Clean + Modern

Incompatible:
❌ Photorealistic + Cartoon
❌ Medieval + Futuristic (需要特殊处理)
❌ Watercolor + Hyperrealistic

[主风格] + [辅助风格] + [技术实现] + [情绪调整]

示例:
"Steampunk (主风格)
+ Art Nouveau decorative elements (辅助风格)
+ 3D render, octane (技术实现)
+ warm, nostalgic atmosphere (情绪调整)"

interface PromptTemplate {
  category: string;
  template: string;
  variables: string[];
  examples: string[];
}

const templates: PromptTemplate[] = [
  {
    category: "Portrait Photography",
    template: `Photorealistic portrait of {subject},
      {lens} lens, {aperture},
      {lighting} lighting,
      {depth_of_field},
      {additional_details},
      professional photography, ultra-detailed, 8k`,
    variables: ["subject", "lens", "aperture", "lighting", "depth_of_field", "additional_details"],
    examples: [
      "subject: a young woman",
      "lens: 85mm",
      "aperture: f/1.8",
      "lighting: natural window",
      "depth_of_field: shallow depth of field",
      "additional_details: detailed skin texture, expressive eyes"
    ]
  },
  // ... 更多模板
];

"Product photography of {product},
{lens} lens, {aperture},
studio lighting with {light_setup},
{background} background,
{special_effects},
commercial photography,
ultra-detailed, 8k"

"{art_medium} illustration,
abstract {subject},
{color_palette} color scheme,
{composition} composition,
{mood} atmosphere,
{additional_style}"

Quality Issues:
"(worst quality:1.4), (low quality:1.4),
blurry, out of focus, distorted,
pixelated, jpeg artifacts"

Anatomy Issues:
"bad anatomy, deformed, disfigured,
extra limbs, extra fingers, missing limbs,
poorly drawn hands, poorly drawn face"

Unwanted Elements:
"watermark, signature, text, logo,
username, artist name, copyright"

Style Issues:
"cartoon, anime, 3d render (如果要求2D),
oversaturated, overexposed, underexposed"

"bad anatomy, deformed face,
asymmetric eyes, crooked teeth,
double chin, bad proportions,
extra fingers, mutated hands"

"distorted perspective, crooked lines,
unrealistic proportions, floating objects,
inconsistent lighting, blurry details"

1. 准备参考图
   ↓
2. 选择合适的 ControlNet 类型
   ↓
3. 调整 ControlNet 权重 (0.5-1.5)
   ↓
4. 编写 Prompt
   ↓
5. 生成并微调

ControlNet: OpenPose
Reference: 舞蹈姿势图
Weight: 1.0

Prompt:
"Professional dancer in elegant pose,
flowing dress, stage lighting,
dramatic atmosphere,
photorealistic, 8k"

Result: 保持姿势，改变服装和风格

基础语法:
(keyword)       # 权重 1.1
((keyword))     # 权重 1.21
[keyword]       # 权重 0.9
(keyword:1.5)   # 精确权重 1.5
(keyword:0.8)   # 精确权重 0.8

组合使用:
"(detailed face:1.3), (expressive eyes:1.2),
natural makeup, (soft smile:1.1),
(photorealistic:1.4)"

问题: 主体不够突出
解决: 增加主体权重
"(main subject:1.5), background"

问题: 某元素过于强烈
解决: 降低权重或添加到 Negative
"subject, (unwanted element:0.5)"
或
Negative: "(unwanted element:1.3)"

问题: 风格不够明显
解决: 增加风格词权重
"subject, (cyberpunk style:1.4),
(neon lights:1.3)"

interface BatchConfig {
  basePrompt: string;
  variations: string[];
  parameters: {
    steps: number;
    cfg: number;
    sampler: string;
  };
  count: number;
}

async function batchGenerate(config: BatchConfig): Promise<void> {
  for (const variation of config.variations) {
    const prompt = `${config.basePrompt}, ${variation}`;
    
    for (let i = 0; i < config.count; i++) {
      await generateImage({
        prompt,
        ...config.parameters,
        seed: Math.floor(Math.random() * 1000000)
      });
    }
  }
}

// 使用示例
const config: BatchConfig = {
  basePrompt: "Portrait of a person",
  variations: [
    "natural lighting",
    "dramatic lighting",
    "soft lighting",
    "golden hour lighting"
  ],
  parameters: {
    steps: 30,
    cfg: 7.5,
    sampler: "DPM++ 2M Karras"
  },
  count: 3
};

await batchGenerate(config);

interface ABTest {
  promptA: string;
  promptB: string;
  parameters: Record<string, unknown>;
  iterations: number;
}

async function runABTest(test: ABTest): Promise<void> {
  const resultsA: string[] = [];
  const resultsB: string[] = [];
  
  for (let i = 0; i < test.iterations; i++) {
    const seed = Math.floor(Math.random() * 1000000);
    
    // 测试 A
    const imageA = await generateImage({
      prompt: test.promptA,
      seed,
      ...test.parameters
    });
    resultsA.push(imageA);
    
    // 测试 B (相同 seed)
    const imageB = await generateImage({
      prompt: test.promptB,
      seed,
      ...test.parameters
    });
    resultsB.push(imageB);
  }
  
  // 对比分析
  console.log("A vs B comparison ready for review");
}