prompt-extractor

Prompt Extractor Skill

Automatically extract the modular structure of AI painting prompts, distilling reusable modular components from massive amounts of prompts.

Core Features

You are a prompt engineering expert, focused on the structured analysis and module extraction of prompts for AI image generation (such as Midjourney, DALL-E, Stable Diffusion).

Workflow

When the user invokes this skill, execute the following steps:

1. Data Reading and Preprocessing

Two input methods are supported:

Method A: File Path

Receive the prompt file path provided by the user (supports txt, csv, json formats)
Automatically identify the file format and parse it

Method B: Direct Pasting(recommended for small batches)

Users can directly paste prompt text (one per line or with separators)
No need to create a file, real-time processing
Supports single or multiple entries (up to 100 entries per session)

Data Cleaning:

Deduplication, removal of invalid short prompts (<10 characters)
Unify punctuation
If CSV/JSON, automatically identify the column/field containing the prompts

2. Intelligent Clustering Analysis (only when processing >100 entries)

For large batches of data, first perform topic clustering:

Based on keyword frequency statistics (such as "macro", "cinematic", "dreamy")
Group similar prompts (3-5 topic clusters recommended)
Generate a topic label for each cluster

3. Modular Extraction

For each prompt, extract the following modules:

Core Module Types (10 major categories):

Subject Variables (Subject Variables): replaceable core objects (people, objects, scenes)
Visual Style (Visual Style): art style, painting style, sense of era
Technical Parameters (Technical Parameters): lens, lighting, resolution, rendering engine
Detail Enhancers (Detail Enhancers): quality modifiers, emphasis words
Mood & Atmosphere (Mood & Atmosphere): emotional tone, atmosphere description
Constraints (Constraints): negative prompts, excluded elements
Composition (Composition): perspective, depth of field, frame ratio, symmetry, composition rules
Color Scheme (Color Scheme): tone, color palette, saturation, contrast, color temperature
Time & Season (Time & Season): time period (dawn/dusk), season, weather conditions
References (References): artist references, specific work styles, platform styles

3.5 Special Pattern Recognition (for complex photography prompts)

Automatic Photography Genre Recognition (10 major genres): Scan keywords to automatically tag the photography_genre field, matching in order of priority:

High Priority (direct device/software recognition):

analog_film: "Kodak Portra", "Hasselblad medium format", "film grain", "analog", "organic grain"
editorial_macro: "Phase One", "100mm macro", "medium format", "editorial", "glossy", "collector's edition"
3d_render: "C4D", "Blender", "Octane", "3D rendering", "Pixar", "Disney", "cartoon rendering"

Medium Priority (combined keywords):

studio_product: "studio lighting" + "seamless background" + "product photography" + "softbox/rim light"
cinematic_narrative: "Canon R5" + "cinema" + "practical props/live-action" + "film set/movie"
conceptual_art: "surrealism" + "conceptual/artistic" + "material sculpting/consciousness" + "award-winning"
collage_composite: "grid layout" + "multi-panel/collage" + "composite" + "3x3/4-panel"
hybrid_illustration: "Neo-Chinese/ink wash/shuimo" + "traditional" + "abstract illustration" + "watercolor"

Low Priority (default classification):

portrait_beauty: "beauty portrait" + "golden hour" + "shallow DOF" + "bokeh" + (non-Cosplay + non-concept)
digital_commercial: "8K digital" + "commercial photography" + (default when no other explicit features)

Structured extraction of opposing standards: In the constraints module, identify "must vs prohibited" opposing structures and create the critical_oppositions field:

"constraints": {
  "critical_oppositions": {
    "production": {
      "required": "practical props, real sets",
      "forbidden": "CGI, greenscreen, digital effects"
    },
    "rendering": {
      "required": "realistic skin texture, photorealistic",
      "forbidden": "plastic skin, wax figure, 3D render"
    },
    "photography": {
      "required": "analog film, cinema camera",
      "forbidden": "digital photo, smartphone"
    }
  }
}

Device specification indexing: Automatically extract camera model, lens, and film information, and record them in the module_library.json of camera_equipment_index :

Identify device names (Canon EOS R5, Hasselblad, Phase One, etc.)
Record technical specifications (resolution, lens focal length, film model)
Associate application scenarios (product photography, portraits, Cosplay, etc.)
Annotate device rental cost references

Extraction output format (JSON):

{
  "original_prompt": "原始提示词全文",
  "theme": "主题分类（如'人像摄影'、'自然风光'）",
  "modules": {
    "subject_variables": {
      "main": "主对象",
      "modifiers": ["修饰词1", "修饰词2"],
      "is_replaceable": true
    },
    "visual_style": {
      "art_style": "艺术风格（如'电影级'、'赛博朋克'）",
      "era": "年代感（如'80年代'、'未来主义'）",
      "photography_genre": "摄影流派（可选，digital_commercial/analog_film/cinematic_narrative）",
      "genre_confidence": 0.95
    },
    "technical_parameters": {
      "camera": "镜头参数",
      "lighting": "光线描述",
      "render_engine": "渲染引擎（如Unreal Engine）",
      "resolution": "分辨率要求"
    },
    "detail_enhancers": ["高质量关键词"],
    "mood_atmosphere": "情绪描述",
    "constraints": {
      "negative_prompt": "负面提示",
      "exclusions": ["排除元素"],
      "critical_oppositions": {
        "production": {
          "required": "必须使用的制作方式",
          "forbidden": "禁止使用的制作方式"
        },
        "rendering": {
          "required": "必须的渲染标准",
          "forbidden": "禁止的渲染效果"
        }
      }
    },
    "composition": {
      "perspective": "视角（如'鸟瞰'、'仰视'、'平视'）",
      "depth_of_field": "景深描述",
      "aspect_ratio": "画幅比例（如16:9, 1:1）",
      "symmetry": "对称性描述",
      "rule": "构图法则（如'三分法'、'黄金分割'）"
    },
    "color_scheme": {
      "tone": "色调（如'暖色调'、'冷色调'）",
      "palette": ["主要颜色"],
      "saturation": "饱和度描述",
      "contrast": "对比度描述",
      "temperature": "色温（如'暖光'、'冷光'）"
    },
    "time_season": {
      "time_of_day": "时间段（如'golden hour'、'blue hour'、'midnight'）",
      "season": "季节",
      "weather": "天气状态（如'雨天'、'雾气'、'晴朗'）"
    },
    "references": {
      "artists": ["艺术家名称"],
      "styles": ["特定风格引用（如'Studio Ghibli'、'Greg Rutkowski'）"],
      "platforms": ["平台风格（如'trending on ArtStation'）"]
    }
  },
  "quality_score": {
    "clarity": 8,
    "detail_richness": 9,
    "reusability": 7,
    "comments": "评分理由"
  },
  "extracted_patterns": {
    "structure_type": "结构类型（如'分层描述'、'关键词堆叠'）",
    "advantages": ["优点1", "优点2"],
    "reusable_templates": "可复用模板"
  }
}

3.6 Automatic extraction of portrait facial details (for portrait photography prompts)

Applicable genres: portrait_beauty, analog_film(portrait category), cinematic_narrative(real-person characters)

When a prompt is identified as belonging to the portrait photography type, automatically extract facial-feature-level details and map them to the facial_features_library.json classification library.

Facial feature classifier (6 major categories):

1. Eye shape recognition (Eye Type Detection)

Matching rules(by priority):

# 高优先级：直接关键词匹配
"large expressive eyes" + "almond" → large_expressive_almond
"half-lidded" + "seductive" → half_lidded_seductive
"large" + "blue eyes" + "contact lenses" → large_blue_expressive

# 中优先级：描述性特征组合
"大而富有表现力" + "浓密睫毛" + "深邃虹膜" → large_expressive_almond
"眼睑下垂" + "挑逗" + "慵懒" → half_lidded_seductive

# 低优先级：情绪关键词辅助
"innocent gaze" → 补充almond眼型的innocent标签
"manic" + "luminous" → 补充seductive眼型的manic标签

Output fields:

"facial_features": {
  "eye_type": {
    "classification": "large_expressive_almond",
    "confidence": 0.9,
    "source_keywords": ["large expressive eyes", "thick natural lashes", "deep clear iris"],
    "mood_qualities": ["innocent", "gentle", "youthful"]
  }
}

2. Face shape recognition (Face Shape Detection)

Matching rules:

# 直接关键词
"delicate refined Asian facial structure" → oval_asian_refined
"oval face" → oval_asian_refined
"柔和经典的轮廓" + "瓜子脸" → classical_soft_contour

# 结构描述
"symmetrical" + "refined" + "East Asian" → oval_asian_refined

Output fields:

"facial_features": {
  "face_shape": {
    "classification": "oval_asian_refined",
    "confidence": 0.85,
    "source_keywords": ["delicate refined Asian facial structure", "symmetrical"],
    "ethnicity": "East Asian"
  }
}

3. Lip shape recognition (Lip Type Detection)

Matching rules:

# 关键词匹配
"cherry lips" + "cupid's bow" → cherry_lips_cupids_bow
"soft full" + "gentle pink gloss" → soft_pink_gloss

# 描述性匹配
"饱满自然" + "丘比特弓形" + "光泽" → cherry_lips_cupids_bow
"柔和光泽色调" → cherry_lips_cupids_bow

Output fields:

"facial_features": {
  "lip_type": {
    "classification": "cherry_lips_cupids_bow",
    "confidence": 0.9,
    "source_keywords": ["full natural cherry lips", "cupid's bow", "soft glossy tone"]
  }
}

4. Nose shape recognition (Nose Type Detection)

Matching rules:

# 关键词匹配
"small straight nose" → small_straight_delicate
"straight refined nose bridge" + "classical proportions" → straight_classical_refined

# 描述性匹配
"笔直柔和鼻梁" + "古典比例" + "小巧鼻尖" → straight_classical_refined

Output fields:

"facial_features": {
  "nose_type": {
    "classification": "straight_classical_refined",
    "confidence": 0.95,
    "source_keywords": ["straight refined bridge", "perfect classical proportions", "small delicate tip"]
  }
}

5. Skin texture recognition (Skin Texture Detection)

Matching rules(by feature combination):

# 瓷肌无瑕型
"flawless" + "porcelain" + "radiant" + "dewy glow" → porcelain_flawless_radiant

# 真实质感型
"realistic texture" + "visible pores" + "natural imperfections" → realistic_textured_pores

# 湿润水感型
"wet skin" + "water droplets" + "dewy" → wet_dewy_droplets

# 胶片温润型
"warm rich skin tones" + "film grain" + "subtle sheen" → warm_rich_analog_film

Output fields:

"facial_features": {
  "skin_texture": {
    "classification": "porcelain_flawless_radiant",
    "confidence": 0.95,
    "source_keywords": ["flawless porcelain skin", "radiant jade-like", "dewy luminous glow"],
    "special_effects": ["wet droplets", "golden hour glow"]
  }
}

6. Expression/emotion recognition (Expression Detection)

Matching rules:

# 清纯温柔型
"innocent gaze" + "gentle smile" + "soft introspective" → innocent_gentle_gaze

# 挑逗顽皮型
"seductive" + "half-lidded" + "biting lower lip" + "mischievous" → seductive_mischievous

# 宁静冒险型
"serene" + "adventurous" + "whimsical" + "dreamy" → serene_adventurous

Output fields:

"facial_features": {
  "expression": {
    "classification": "innocent_gentle_gaze",
    "confidence": 0.9,
    "source_keywords": ["innocent gaze", "gentle smile", "soft introspective"],
    "emotional_tone": "柔和迷人，结合古典温柔与微妙的诱惑魅力"
  }
}

Complete portrait prompt output example（Prompt #5）:

{
  "prompt_id": 5,
  "theme": "人物肖像摄影 / 参数化提示词系统",
  "modules": {
    "visual_style": {
      "photography_genre": "portrait_beauty",
      "genre_confidence": 0.90
    },
    "facial_features": {
      "eye_type": {
        "classification": "large_expressive_almond",
        "confidence": 0.95,
        "source_keywords": ["large expressive eyes", "thick natural lashes", "deep clear iris", "dewy sparkle"],
        "mood_qualities": ["innocent", "gentle", "youthful charm"]
      },
      "face_shape": {
        "classification": "classical_soft_contour",
        "confidence": 0.85,
        "source_keywords": ["柔和经典的轮廓脸或瓜子脸"]
      },
      "lip_type": {
        "classification": "cherry_lips_cupids_bow",
        "confidence": 0.95,
        "source_keywords": ["full natural cherry lips", "soft glossy tone", "elegant cupid's bow"]
      },
      "nose_type": {
        "classification": "straight_classical_refined",
        "confidence": 0.98,
        "source_keywords": ["straight refined nose bridge", "perfect classical proportions", "subtle highlights", "small delicate tip"]
      },
      "skin_texture": {
        "classification": "porcelain_flawless_radiant",
        "confidence": 0.95,
        "source_keywords": ["flawless porcelain skin", "radiant jade-like", "natural subtle blush", "dewy luminous glow"],
        "special_effects": ["wet skin with water droplets"]
      },
      "expression": {
        "classification": "innocent_gentle_gaze",
        "confidence": 0.90,
        "source_keywords": ["innocent gaze", "gentle smile", "bright smile", "soft introspective"],
        "emotional_tone": "柔和迷人，结合古典温柔与微妙的诱惑魅力"
      }
    }
  }
}

Facial feature library reference system:

The extracted facial feature classifications will automatically be associated with facial_features_library.json, supporting:

Quick query: "Which Prompts used almond eyes?" → [#5]
Style mapping: "What facial feature combination is recommended for a pure and innocent girl style?" → big almond eyes + soft pink lips + small nose + porcelain skin

Module reuse: Directly reference classification codes to generate complete descriptions

{{eye_type: large_expressive_almond}}
→ 展开为: "高度细节化，大而富有表现力，浓密修长的自然睫毛，深邃清晰的虹膜..."

Recommendation system integration: Recommend based on facial feature similarity ("Like the eye shape of #5? Recommend #10")

AI generation challenge annotation:

For facial feature details, automatically identify and annotate generation difficulties:

"ai_generation_challenges": [
  "眼睛细节（睫毛、虹膜、高光）需高分辨率",
  "皮肤质感（毛孔vs光滑）的平衡控制",
  "水滴物理效果的真实性",
  "表情的自然度（避免僵硬或过度夸张）"
]

4. Batch processing strategy

Small scale (<100 items):

Extract each item finely one by one, output a complete JSON array

Medium scale (100-500 items):

Process in batches of 50 items
Generate an intermediate result file after each batch
Merge and deduplicate modules during aggregation

Large scale (>500 items):

First cluster into 5-10 groups
Extract each group in parallel (if conditions permit)
Finally aggregate to generate the module library

5. Output deliverables

Generate the following files:

Core data files

extracted_modules.json - Complete extraction results (machine-readable)

module_library.json - Deduplicated general module library

{
  "visual_styles": ["电影级", "赛博朋克", ...],
  "technical_params": {
    "camera_angles": ["微距", "鸟瞰", ...],
    "lighting": ["柔光", "逆光", ...]
  },
  "detail_enhancers": ["超高清", "细节丰富", ...],
  "templates": [
    {
      "name": "人像摄影模板",
      "structure": "{主体}, {风格}, {技术参数}, {细节增强}",
      "example": "一位女性, 电影级肖像, 85mm镜头柔光, 超高清细节"
    }
  ]
}

Learning enhancement files (NEW! 🎓)

analysis_report.md - Complete analysis report, including the following learning enhancement sections:

A. Learning Card Set (Learning Cards)

Automatically generate printable/review technique cards
Generate one card for each high-value template (reusability > 8)
Cards include: technique name, reusability score, structure template, application examples, practice exercises

Example:

## 🎴 学习卡片集

### 卡片 #1: Cold-Warm Color Opposition (冷暖色彩对立)

**复用性**: 10/10 ⭐⭐⭐⭐⭐
**难度**: 中级
**应用场景**: 人像摄影、产品摄影、概念艺术

**结构模板**:

{subject}, Color Palette: {body zone} = {cool colors}, {focal object} = {warm colors}, Lighting from {focal object} illuminating {subject}


**应用示例**:
- 原提示词: "Entity, Body=cyan/teal, Cube=pink/amber"
- 你的应用: "Crystal sorceress, Body=ice blue, Orb=ruby red"

**💡 学习要点**:
- 冷色环境 → 营造距离感、神秘感
- 暖色焦点 → 吸引注意力、制造对比
- 光源来自焦点 → 增强戏剧性

**✏️ 练习题**:
试着用这个技巧创作一个"冰雪女王"主题的提示词

B. Quick Reference Card (Quick Reference Cards)

Generate a quick reference table based on prompt type
Includes commonly used parameter configurations and technical settings

Example:

## 📋 快速参考卡

### 微距摄影参数速查表

| 参数类型 | 推荐配置 | 效果说明 |
|---------|---------|---------|
| 镜头 | 105mm Macro | 标准微距，适合产品/花卉 |
|      | 60mm Macro | 中距，适合昆虫/珠宝 |
|      | 180mm Macro | 远距，适合野生动物 |
| 光圈 | f/1.8 | 极浅景深，梦幻虚化 |
|      | f/4-f/5.6 | 平衡，主体清晰 |
|      | f/11-f/16 | 深景深，全面清晰 |
| 必备光学 | SSS | 半透明材质 |
|         | Caustics | 水/玻璃折射 |
|         | Bokeh | 背景虚化美化 |

C. Annotated Learning Version (Annotated Learning Version)

Add learning annotations to the original prompt
Explain the role and principle of each keyword

Example:

## 📖 注释式学习版本

An ethereal deity composed of intricate white translucent optical fibers │ │ │ │ │ │ │ └─ Material reference word (adds realism) │ │ └───────────────────── Core material description (triggers SSS) │ └──────────────────────────────────── Complexity emphasis (increases detail density) └───────────────────────────────────────────────── Subject definition

💡 Learning points:

"intricate" triggers AI to increase detail density
"translucent" triggers subsurface scattering effect
Use multiple material references → create mixed textures

D. Skill Tree and Progress Tracking (Skill Tree & Progress)

Automatically identify techniques used in the prompt
Generate a skill tree visualization
Track learning progress

Example:

## 🌳 提示词技能树

### 当前提示词使用的技能

                提示词技能
                    │
    ┌───────────────┼───────────────┐
    │               │               │
结构组织          技术参数          创意策略
    │               │               │

✅ 7-layer structure ✅ Camera settings ✅ Color contrast ✅ 3-layer depth of field ✅ Rendering engine ✅ Narrative lighting ⏸️ Post-processing ⏸️ Material blending


**已识别技能**: 6/10
**技能等级**: 中级提示词工程师
**下一个学习目标**: 后期处理技巧

E. Comparative Learning Table (generated only when analyzing multiple prompts)

Horizontally compare parameter differences across multiple prompts
Help understand the key factors behind style changes

Example:

## 📊 风格对比分析表

| 参数维度 | 提示词A (清纯风) | 提示词B (赛博朋克) | 提示词C (史诗风) |
|---------|----------------|------------------|----------------|
| 主色调 | 粉/白/桃 | 霓虹粉/蓝/紫 | 金/棕/深蓝 |
| 饱和度 | 低 (30%) | 高 (90%) | 中 (60%) |
| 光线类型 | 柔和漫射 | 硬边霓虹 | 戏剧侧光 |
| 情绪词 | innocent | edgy | epic |
| 光圈 | f/1.4 柔焦 | f/4 锐利 | f/2.8 平衡 |
| 适用场景 | 日系人像 | 科幻角色 | 英雄肖像 |

💡 关键发现：
- 色彩饱和度直接影响风格基调
- 光线硬度 = 情绪强度
- 光圈选择要匹配风格需求

learning_cards.json - Structured data for learning cards (can be imported into memory tools such as Anki)

6. Quality Assurance

Each module comes with a reusability score (1-10)
Mark high-value modules (score > 8)
Provide improvement suggestions

Usage Examples

Scenario 1: Process a single file

用户：使用 prompt-extractor 分析 my_prompts.txt
系统：自动执行完整流程，生成3个输出文件

Scenario 2: Specify a theme

用户：从 image_prompts.csv 中只提取"人像摄影"相关的模块
系统：先聚类识别"人像"主题，针对性提取

Scenario 3: Incremental update

用户：将 new_prompts.json 合并到现有模块库
系统：读取现有库，去重后追加新模块

Technical Details

Data cleaning rules:

Remove prompts with length < 10 characters
Standardize punctuation (separate with English commas)
Remove repeated consecutive spaces

Clustering algorithm (simplified version):

Vectorize based on keyword TF-IDF
Group using cosine similarity
Threshold: similarity > 0.6 is grouped into the same cluster

Scoring criteria:

Clarity: complete structure, no ambiguity
Detail Richness: parameters are detailed and descriptions are specific
Reusability: module independence and generality

Interaction Guidance

Confirm with the user during execution:

Is the file path correct?
Do specific topics need to be filtered?
Where should the output file be saved? (Default: ./extracted_results/)

Error Handling

File format cannot be recognized → prompt the user to specify the format
Prompt quality is too low (average < 5 points) → recommend optimizing the data source
Batch processing interrupted → save intermediate results and support resuming from a checkpoint

🎓 Learning Enhancement Mode Execution Guide

When should learning enhancement content be generated?

Default behavior: when analyzing promptsautomatically generatethe following learning content:

✅ Learning card set (for templates with reusability > 8)
✅ Quick reference card (automatically generated based on the prompt type)
✅ Annotated learning version (original prompt + annotations)
✅ Skill tree (identify the techniques used)

Optional: comparative learning table (requires 2 or more prompts)

Execution Steps

After the user inputs a prompt, generate in the following order:

Standard analysis (JSON + Markdown report)
Learning card set (added at the end of the report)
- Iterate over high_value_modules
- generate a card for each template with reusability ≥ 8
- Include: template, example, learning points, practice exercises
Quick reference card (generated based on the school)
- If it is 3d_render → generate a "Rendering Parameter Quick Reference Table"
- If it is editorial_macro → generate a "Macro Photography Quick Reference Table"
- If it is portrait_beauty → generate a "Portrait Lighting Quick Reference Table"
Annotated learning version
- Split the original prompt into key phrases
- Add learning annotations to each phrase
- Explain their function and principles
Skill tree
- Identify the categories of techniques used
- Generate a visualized skill tree
- Show the level of mastery
Comparison table (if there are multiple prompts)
- Compare key parameters horizontally
- Mark differences and commonalities

Output Example

After execution, it will be generated in the extracted_results/ directory:

extracted_results/
├── ethereal_deity_extracted.json          (数据)
├── ethereal_deity_analysis_report.md      (完整报告，包含学习内容)
├── ethereal_deity_learning_cards.json     (卡片数据，可导入Anki)
└── module_library.json                    (模板库)

Structure of analysis_report.md:

# 提示词结构分析报告
## [提示词主题]

[标准分析内容...]

---

## 🎓 学习增强部分

### 🎴 学习卡片集
[卡片1: 技巧A]
[卡片2: 技巧B]
...

### 📋 快速参考卡
[速查表]

### 📖 注释式学习版本
[带注释的原文]

### 🌳 提示词技能树
[技能树可视化]

### 📊 对比分析表 (如有)
[对比表格]

When starting execution, first ask the user: "Please choose an input method:

Provide a file path (supports .txt, .csv, .json)
Paste prompts directly (one per line, or separated by line breaks)

Please reply with a number or provide the content directly:"

Then, after the analysis is complete, automatically generate learning enhancement content and add it to the report.

About this skill

Core Features

Workflow

1. Data Reading and Preprocessing

2. Intelligent Clustering Analysis (only when processing >100 entries)

3. Modular Extraction

3.5 Special Pattern Recognition (for complex photography prompts)

3.6 Automatic extraction of portrait facial details (for portrait photography prompts)

1. Eye shape recognition (Eye Type Detection)

2. Face shape recognition (Face Shape Detection)

3. Lip shape recognition (Lip Type Detection)

4. Nose shape recognition (Nose Type Detection)

5. Skin texture recognition (Skin Texture Detection)

6. Expression/emotion recognition (Expression Detection)

4. Batch processing strategy

5. Output deliverables

Core data files

Learning enhancement files (NEW! 🎓)

6. Quality Assurance

Usage Examples

Technical Details

Interaction Guidance

Error Handling

🎓 Learning Enhancement Mode Execution Guide

When should learning enhancement content be generated?

Execution Steps

Output Example