prompt-generator

name	prompt-generator
description	从分镜表的画面描述生成完整的 AI 生图提示词。核心功能：自动注入角色一致性 Token，确保同一角色跨镜头视觉一致。支持 Midjourney / DALL-E / Stable Diffusion 等平台。

Prompt Generator — 剧照提示词生成器

从分镜表中的画面描述生成可直接使用的 AI 生图提示词。核心机制：自动从角色档案加载一致性 Token，注入到每个包含该角色的剧照提示词中，保证角色在所有镜头中外貌一致。

When to Use This Skill

• 从分镜表生成剧照 (角色镜头) 的 AI 生图提示词
• 从 B-Roll 规划生成补充镜头提示词
• 批量生成一组风格一致的图片提示词
• 需要跨多张图保持角色外貌一致
• 为不同 AI 平台优化提示词格式
• 翻译中文描述为英文提示词

What This Skill Does

1. 角色 Token 注入: 从 ./output/characters/ 加载一致性 Token，自动嵌入提示词
2. 画面→提示词: 将分镜的画面描述转化为结构化提示词
3. 风格一致: 使用共享的风格基底 (Style Foundation) 确保系列一致
4. 平台适配: 为 Midjourney / DALL-E / SD 分别优化
5. 反向提示词: 生成精确的 Negative Prompt
6. 批量输出: 批量生成可直接复制使用的提示词文件

How to Use

从分镜表批量生成

从分镜表生成所有剧照提示词

指定镜头生成

为 S01-C01 和 S02-C03 生成生图提示词

风格参考生成

参考这张图的风格，为分镜表生成提示词: [图片描述/URL]

Instructions

When user requests prompt generation:

1. 加载角色 Token 和风格基底

   ## 角色一致性 Token 加载
   
   **角色档案来源**: `./output/characters/`
   
   | 角色 | Token 状态 | 简短版 Token |
   |------|-----------|-------------|
   | A: [Name] | ✅ 已加载 | `30-year-old asian woman with shoulder-length black hair...` |
   | B: [Name] | ✅ 已加载 | `45-year-old man with short gray hair, square jaw...` |
   
   **风格基底** (Style Foundation):
   

   [art style], [quality: 4K, sharp focus, detailed], [lighting: soft three-point / cinematic / natural], [render: photorealistic / 3D / illustration], [color: warm palette / brand colors #XXX #XXX], [camera: professional photography / cinematic]

   
   > ⚠ 以下所有提示词将自动注入对应角色的 Token
   > ⚠ 风格基底统一应用于所有提示词
   

2. 角色 Token 注入机制

   ## Token 注入规则
   
   ### 提示词组装公式
   
   

   PROMPT = [SCENE/环境描述] + [CHARACTER TOKEN: 角色完整外貌] + [POSE/动作姿态] + [EXPRESSION/表情] + [CAMERA/镜头参数] + [STYLE FOUNDATION/风格基底]

   
   ### 规则
   
   1. **必须注入**: 任何包含角色的剧照，必须完整注入该角色的 Token
   2. **不可修改**: Token 中的面部和体型描述不可修改 (除非换装)
   3. **换装替换**: 如场景需要不同服装，仅替换 Token 中的服装部分
   4. **多角色**: 多个角色出现时，每个角色都注入各自的 Token
   5. **B-Roll**: 不含角色的画面不需要注入 Token
   6. **表情追加**: 在 Token 后追加当前镜头的表情描述
   
   ### 注入示例
   
   **输入** (从分镜表):
   - 分镜: S01-C01
   - 角色: A
   - 姿态: 站在中央偏右, 面对镜头
   - 表情: 自信微笑
   - 场景: 简洁演播室
   - 景别: 中景
   
   **输出** (注入后的提示词):
   

   Clean minimalist studio background, soft gray tones, [30-year-old asian woman with shoulder-length black hair and side bangs, wearing navy blazer over white blouse, wire-rimmed glasses, slim build], standing center-right of frame, facing camera directly, confident warm smile, hands clasped naturally in front, medium shot from waist up, eye-level angle, soft three-point lighting, key light from right, photorealistic, 4K, sharp focus, professional portrait

   
   > 注意: `[30-year-old asian woman...]` 就是从 
   > `./output/characters/[name]/consistency-tokens.md` 
   > 加载的角色 Token — 原封不动嵌入
   

3. 分镜→提示词批量转换

   ## 批量提示词生成
   
   **来源**: `./output/storyboard/photo-gen-tasks.md`
   **总计**: [N] 张剧照
   
   ---
   
   ### Prompt S01-C01: [场景标题] — [角色 A] 自信开场
   
   **Prompt**:
   

   [Complete prompt with character token injected, scene description, pose, expression, camera, style]

   
   **Negative Prompt**:
   

   blurry, low quality, distorted face, extra limbs, bad anatomy, deformed, watermark, text, multiple characters, wrong outfit, extra accessories, wrong hair color, wrong hair style

   
   **参数**:
   - 平台: [Midjourney / SD / DALL-E]
   - 比例: 16:9 (--ar 16:9)
   - 种子: [seed for consistency]
   - 风格: [--style raw / --s 250]
   
   **来源分镜**: S01-C01
   **角色**: [A]
   **用途**: 主镜头剧照
   
   ---
   
   ### Prompt S01-C02: [场景标题] — [角色 A] 好奇提问
   
   [Same structure...]
   
   ---
   
   ### Prompt S01-B01: B-Roll — 科技氛围
   
   **Prompt**:
   

   [No character token — B-Roll without characters] Abstract digital data flow, glowing particles...

   
   [No character token needed for B-Roll]
   
   ---
   
   [Continue for all shots...]
   

4. 平台适配优化

   For Midjourney:

   ### Midjourney Prompt Template
   
   [Subject], [style], [setting], [lighting], [camera angle], [mood] --ar [ratio] --v 6 --style raw --s [stylize value]
   
   **Parameter Guide**:
   - `--ar 16:9`: Aspect ratio for video
   - `--v 6`: Version 6 (latest)
   - `--style raw`: More photographic
   - `--s 250`: Stylize (0-1000, higher = more artistic)
   - `--c 20`: Chaos (0-100, higher = more variation)
   - `--q 2`: Quality (0.25, 0.5, 1, 2)
   
   **Example**:
   

   modern office workspace with natural light streaming through large windows, professional business setting, minimalist design with plants, soft morning light, eye-level view, bright and airy atmosphere, photorealistic, architectural photography style, sharp focus, --ar 16:9 --v 6 --style raw --s 250

   For DALL-E 3:

   ### DALL-E 3 Prompt Template
   
   [Detailed natural language description, longer and more narrative]
   
   **Tips**:
   - Use natural, descriptive language
   - Be specific about style and mood
   - Describe colors, lighting, composition
   - Can be longer than other platforms
   - No need for parameters
   
   **Example**:
   

   A modern office workspace bathed in warm natural morning light streaming through floor-to-ceiling windows. The space features a minimalist aesthetic with clean lines, a sleek wooden desk, and carefully placed green plants adding life to the scene. The lighting creates a soft, inviting atmosphere with gentle shadows. Shot from eye level to capture the balanced composition and professional yet welcoming environment. Photorealistic style with sharp focus and architectural photography aesthetics.

   For Stable Diffusion:

   
   
   [Subject], [detailed description], ([style keywords]), ([quality keywords]), [lighting], [camera]
   
   :
    (keyword): mild emphasis
    ((keyword)): stronger emphasis
    (keyword:1.3): weighted emphasis (1.0-2.0)
   
   :
   

5. Style Consistency Guidelines

   ## Maintaining Visual Consistency
   
   **For Series of Images**:
   
   **Consistent Elements** (keep in all prompts):
   - Color palette: [Specify exact colors]
   - Lighting setup: [Keep lighting description same]
   - Style keywords: [Use same style descriptors]
   - Quality markers: [Same technical terms]
   - Aspect ratio: [Same ratio for all]
   
   **Variable Elements** (change per prompt):
   - Subject/action
   - Specific setting details
   - Camera angle (if varied shots needed)
   
   **Example Series**:
   
   **Base Style Template**:
   

   [Subject], cinematic lighting, warm color palette with dominant blues and oranges, professional photography, shallow depth of field, 35mm lens, golden hour lighting, photorealistic, 8k uhd, sharp focus --ar 16:9

   
   **Prompt 1**: "tech startup founder [BASE STYLE]"
   **Prompt 2**: "modern workspace [BASE STYLE]"
   **Prompt 3**: "digital dashboard closeup [BASE STYLE]"
   
   [Each uses same style foundation but different subject]
   

6. Quality Enhancement Keywords

   ## Quality & Technical Keywords
   
   **Resolution/Detail**:
   - 8k uhd, 4k, ultra high definition
   - highly detailed, intricate details, sharp focus
   - professional quality, studio quality
   
   **Photorealism**:
   - photorealistic, photo-realistic, hyperrealistic
   - realistic lighting, natural lighting
   - professional photography, DSLR
   - shot on [camera model]
   
   **Cinematic**:
   - cinematic lighting, dramatic lighting
   - depth of field, bokeh, shallow focus
   - film grain, 35mm, anamorphic
   - movie still, film screenshot
   
   **Artistic Style**:
   - digital art, concept art, matte painting
   - illustrated, stylized, artistic
   - trending on artstation, award winning
   
   **Lighting Descriptors**:
   - golden hour, blue hour, magic hour
   - soft light, hard light, dramatic light
   - rim light, back light, volumetric light
   - natural light, studio lighting, professional lighting
   
   **Camera/Lens**:
   - wide angle, telephoto, macro
   - 35mm, 50mm, 85mm lens
   - shallow depth of field, f/1.8, bokeh
   - eye level, high angle, low angle, bird's eye view
   

7. Negative Prompt Library

   ## Common Negative Prompts
   
   **Quality Issues**:
   

   low quality, blurry, out of focus, pixelated, jpeg artifacts, low resolution, grainy (unless intended), noisy, distorted

   
   **Unwanted Elements**:
   

   watermark, text, signature, logo (unless wanted), border, frame, username, watermark

   
   **Anatomical Issues** (if people):
   

   deformed, disfigured, bad anatomy, wrong anatomy, extra limbs, missing limbs, floating limbs, disconnected limbs, mutation, mutated, ugly, disgusting, amputation

   
   **Style Issues**:
   

   amateur, unprofessional, bad composition, bad framing, cluttered, messy, chaotic (unless intended)

   
   **Lighting Issues**:
   

   bad lighting, overexposed, underexposed, harsh shadows (unless intended), too dark, too bright, flat lighting (unless intended)

   
   **Artistic Issues**:
   

   cartoon (unless intended), anime (unless intended), sketch (unless intended), painting (unless photo wanted), drawing, illustration (unless intended)

8. Batch Prompt Generation Template

   ## Batch Prompts: [Topic/Series Name]
   
   **Shared Style Foundation**:
   

   [Common elements across all prompts]

   
   **Negative Prompt** (all prompts):
   

   [Shared negative prompt]

   
   ---
   
   ### Image 001: [Description]
   **Full Prompt**: `[Complete prompt]`
   **Use**: [Where in video]
   **Duration**: [How long shown]
   
   ### Image 002: [Description]
   **Full Prompt**: `[Complete prompt]`
   **Use**: [Where in video]
   **Duration**: [How long shown]
   
   ### Image 003: [Description]
   **Full Prompt**: `[Complete prompt]`
   **Use**: [Where in video]
   **Duration**: [How long shown]
   
   [Continue for all needed]
   
   ---
   
   **Generation Order**: [Recommended sequence]
   **Estimated Total Generation Time**: [Estimate]
   **Platform Recommended**: [Best platform for this batch]
   

9. Test and Iterate Template

   ## Prompt Testing Guide
   
   **Initial Generation**:
   - Generate from Prompt 001
   - Review results
   - Note what works and what doesn't
   
   **Refinement**:
   
   **If too generic**:
   - Add more specific details
   - Add style references
   - Increase stylize/creativity
   
   **If too chaotic**:
   - Simplify description
   - Be more specific
   - Lower chaos/variation parameters
   
   **If wrong style**:
   - Add style keywords
   - Reference specific artists/styles
   - Use style parameters
   
   **If wrong composition**:
   - Be explicit about framing
   - Specify camera angle clearly
   - Describe element placement
   
   **Iteration Versions**:
   
   **v1** (Original): `[Prompt]`
   **v2** (Refinement 1): `[Adjusted prompt]`
   **v3** (Refinement 2): `[Further adjusted]`
   **Final**: `[Best performing prompt]`
   

10. Save Prompts

    保存到 ./output/prompts/:

    ./output/prompts/
    ├── prompt-index.md            # 提示词总索引
    ├── style-foundation.md        # 风格基底定义
    ├── character-shots/           # 角色剧照提示词
    │   ├── S01-C01.md
    │   ├── S01-C02.md
    │   └── ...
    ├── broll-shots/               # B-Roll 提示词
    │   ├── S01-B01.md
    │   └── ...
    └── all-prompts.txt            # 所有提示词 (纯文本，便于复制)
    

11. Provide Summary

    ✅ 剧照提示词生成完成！
    
    **总计**: [N] 组提示词
    - 角色剧照: [N] 张 (已注入一致性 Token)
    - B-Roll: [N] 张
    - 三视图: [N] 张
    
    **保存位置**:
    - 详细版: `./output/prompts/character-shots/`
    - 纯文本: `./output/prompts/all-prompts.txt`
    - 风格基底: `./output/prompts/style-foundation.md`
    
    **角色 Token 状态**:
    - [Name A]: ✅ 已注入 [N] 张
    - [Name B]: ✅ 已注入 [N] 张
    
    **Next Steps**:
    1. 批量生图: `使用 batch-image-generator 生成所有剧照`
    2. 调整提示词: `修改 S01-C02 的表情为 [新表情]`
    3. 更换风格: `将风格改为 3D 卡通`
    4. 换装生成: `为角色 A 使用休闲装方案重新生成`
    

Tips

• Token 完整注入: 不要省略角色 Token 的任何部分，完整性 = 一致性
• 风格基底统一: 所有提示词共享相同的 Style Foundation
• 先测后批: 先用 1-2 张测试效果，再批量生成
• 种子一致: 使用相同种子提高同场景的一致性
• 表情单独控制: 表情描述放在 Token 之后，不修改 Token 本身
• 中→英翻译: 分镜表可以是中文，但提示词输出必须是英文

Related Skills

• character-designer — 上游：提供角色一致性 Token
• storyboard-generator — 上游：提供剧照画面描述
• broll-generator — 并行：提供 B-Roll 画面描述
• batch-image-generator — 下游：使用提示词批量生图
• keyframe-extractor — 参考：从参考视频提取风格