image-prompt-engineer

id	image_prompt_engineer
name	image_prompt_engineer
version	2.1
type	agent-skill
languages	en, th
category	image_prompt_generation
execution_mode	enhance-prompt
chainTo	image-creator
isAutoTrigger	true
enabledByDefault	true
priority	50
triggerPatterns	["สร้างพรอมต์|เขียนพรอมต์|แต่งพรอมต์|enhance prompt|image prompt|prompt สำหรับภาพ","generate.*image.*prompt|พรอมต์.*ภาพ|prompt.*รูป|create.*prompt.*image","create image prompt|write image prompt|enhance image prompt|generate image prompt|create prompt|write prompt","img prompt|สร้าง prompt|เขียน prompt"]
description	Auto-imported from skills/image_prompt_engineer
icon	sparkles
tags	[]
auto_trigger	true
trigger_patterns	["สร้างพรอมต์|เขียนพรอมต์|แต่งพรอมต์|enhance prompt|image prompt|prompt สำหรับภาพ","generate.*image.*prompt|พรอมต์.*ภาพ|prompt.*รูป|create.*prompt.*image","create image prompt|write image prompt|enhance image prompt|generate image prompt|create prompt|write prompt","img prompt|สร้าง prompt|เขียน prompt"]
enabled_by_default	true
credit_multiplier	1
strict_provider_pin	false

Image Prompt Engineer (v2.1)

🆕 v2.1 Update: Hallucination Control

• ✅ Prevents nationality/ethnicity hallucination — Prevents adding unspecified nationality data
• ✅ Auto-correction — Automatic correction (e.g. "Korean fashion" → "modern fashion")
• ✅ Warnings in output — Alerts when hallucination is detected

🎯 Purpose

Create comprehensive, clear "prompts for AI image generation systems" that support all image generation modes:

✅ Supported Modes (New in v2.1!)

1. Text-to-Image — Generate images from text descriptions
2. Image-to-Image — Transform images from reference images
3. Inpaint — Edit only selected areas (Text-based masking)
4. Outpaint — Expand images beyond the original frame
5. Variation — Create variations from an existing image

🌟 Key Features

• Default mode emphasizes realism (can be changed with Style/VFX)
• Supports text on image (Typography) with selectable categories/styles
• Supports multiple reference images (assign roles to each image)
• Text-based Masking — Specify edit areas using natural language
• Platform-specific Output — Adjust prompts per platform
• Advanced Controls — Fine-tune advanced parameters

---

📊 What's New in v2.1

🚀 Major Features

1. Generation Mode Selection

Clearly specify the image generation mode:

{
  "generation_mode": "text_to_image" | "image_to_image" | "inpaint" | "outpaint" | "variation"
}

2. Text-based Inpainting

Edit only the desired areas using natural language:

{
  "generation_mode": "inpaint",
  "edit_mask": {
    "type": "prompt_based",
    "segment_prompt": "sky",
    "preserve_areas": ["foreground", "people"]
  }
}

3. Outpainting Support

Expand images in all directions:

{
  "generation_mode": "outpaint",
  "outpaint_config": {
    "expand_left": 256,
    "expand_right": 256,
    "expand_top": 128,
    "expand_bottom": 128
  }
}

4. Advanced Parameters

Fine-grained control over image generation:

{
  "advanced_params": {
    "denoising_strength": 0.75,
    "guidance_scale": 7.5,
    "steps": 50,
    "seed": 123456,
    "sampler": "dpm_2m_karras"
  }
}

5. ControlNet & IP-Adapter Support

{
  "controlnet": {
    "enabled": true,
    "type": "pose",
    "weight": 1.0
  },
  "ip_adapter": {
    "enabled": true,
    "mode": "style",
    "weight": 0.6
  }
}

6. Platform Selection

Adjust prompts to suit the target platform:

{
  "target_platform": "stable_diffusion" | "midjourney" | "dall_e_3" | "gemini_imagen" | "flux" | "firefly"
}

---

📋 Input Schema

Required Fields

{
  "request": "Description of what you want" // The only required field!
}

Core Fields (All have defaults)

{
  "generation_mode": "text_to_image",  // default
  "background_type": "normal",         // normal | green_screen | blue_screen | transparent
  "task": "final_prompt",              // default
  "detail_level": "standard",          // compact | standard | full
  "languages": "en",                   // en | th
  "aspect_ratio": "9:16",              // 7 options available
  "aspect_ratio_custom": "",           // e.g. "5:4"
  "style": "photorealistic",           // 151+ styles
  "target_platform": "generic"         // 7 platforms
}

Image-to-Image Fields

{
  "reference_images": [
    {
      "role": "primary_subject" | "outfit" | "product" | "location_background" | ...,
      "notes": "Additional description"
    }
  ],
  "identity_lock": "none" | "soft_lock_person" | "strict_lock_product",
  "realistic_skin": false
}

Inpainting Fields

{
  "edit_mask": {
    "type": "prompt_based",              // or ai_segment, rectangle, brush
    "segment_prompt": "sky",             // "sky", "background", "the woman's dress"
    "preserve_areas": ["face", "hands"], // Areas to preserve
    "feather": 10,                       // Edge softness (px)
    "invert": false                      // Invert mask
  }
}

Outpainting Fields

{
  "outpaint_config": {
    "expand_left": 0,      // px
    "expand_right": 0,     // px  
    "expand_top": 0,       // px
    "expand_bottom": 0,    // px
    "blend_width": 64,     // px (blend zone)
    "match_style": true    // Match original style
  }
}

Advanced Parameters

{
  "advanced_params": {
    "denoising_strength": 0.75,  // 0-1 (img2img)
    "guidance_scale": 7.5,        // CFG: 1-30
    "steps": 50,                  // sampling steps: 1-150
    "seed": -1,                   // -1 = random
    "sampler": "dpm_2m_karras",   // euler_a, ddim, etc.
    "clip_skip": 1                // 1-12
  }
}

ControlNet Configuration

{
  "controlnet": {
    "enabled": false,
    "type": "canny",              // depth, pose, normal, scribble, mlsd, lineart, softedge
    "weight": 1.0,                // 0-2
    "guidance_start": 0.0,        // 0-1
    "guidance_end": 1.0           // 0-1
  }
}

IP-Adapter Configuration

{
  "ip_adapter": {
    "enabled": false,
    "mode": "style",              // content, face, composition
    "weight": 0.6,                // 0-2
    "start_step": 0.0,            // 0-1
    "end_step": 1.0               // 0-1
  }
}

VFX Effects

{
  "vfx": {
    "effects": [
      "light_volumetric_lighting",
      "atmospheric_mist"
    ],
    "effects_custom": ["custom effect description"]
  }
}

Typography (Text-on-Image)

{
  "text_on_image": false,
  "headline": "Main text",
  "body_text": "Supporting text",
  "typography": {
    "font_personality": ["modern_clean"],
    "composition_style": ["centered_layout"],
    "mood_tone": ["minimal_and_calm"],
    "color_direction": ["monochrome"],
    "text_effects": ["drop_shadow"],
    "use_case_templates": ["poster_typography"],
    "modern_trend_packs": ["korean_clean_typography"],
    "layout_add_ons": ["with_shapes"]
  }
}

---

📤 Output Schema

{
  "prompt": "Main prompt in the selected language",
  "avoid": ["List of things to avoid"],
  "detail_level": "standard",
  "task": "final_prompt",
  "generation_mode": "text_to_image",
  "target_platform": "generic",
  "parameters": {
    "aspect_ratio": "9:16",
    "generation_mode": "text_to_image",
    "denoising_strength": 0.75,  // if applicable
    "cfg_scale": 7.5,            // if applicable
    "steps": 50                  // if applicable
  },
  "breakdown": {               // if detail_level = full
    "generation_mode": "...",
    "subject": "...",
    "style": "..."
  }
}

---

💡 Usage Examples

Example 1: Text-to-Image (Simple)

{
  "request": "Beautiful woman standing in a flower garden in the morning"
}

✅ Uses all defaults: photorealistic style, 9:16 aspect ratio, standard detail

Example 2: Image-to-Image (Style Transfer)

{
  "request": "Transform into an oil painting",
  "generation_mode": "image_to_image",
  "reference_images": [
    {
      "role": "primary_subject",
      "notes": "Keep the original main composition"
    }
  ],
  "style": "oil_painting",
  "advanced_params": {
    "denoising_strength": 0.8
  }
}

Example 3: Inpainting (Replace Background)

{
  "request": "Change the background to a beach at sunset",
  "generation_mode": "inpaint",
  "reference_images": [
    {
      "role": "primary_subject",
      "notes": "Keep the person in the original image 100%"
    }
  ],
  "edit_mask": {
    "type": "prompt_based",
    "segment_prompt": "background",
    "preserve_areas": ["person", "clothing"],
    "feather": 20
  },
  "style": "golden_hour_cinematic",
  "identity_lock": "soft_lock_person"
}

Example 4: Outpainting (Expand Canvas)

{
  "request": "Expand the image in all directions, show more of the surrounding scene",
  "generation_mode": "outpaint",
  "reference_images": [
    {
      "role": "primary_subject",
      "notes": "Original image to expand"
    }
  ],
  "outpaint_config": {
    "expand_left": 256,
    "expand_right": 256,
    "expand_top": 128,
    "expand_bottom": 128,
    "blend_width": 64,
    "match_style": true
  }
}

Example 5: Advanced - ControlNet + Style

{
  "request": "Create an image in the same pose but in cartoon style",
  "generation_mode": "image_to_image",
  "reference_images": [
    {
      "role": "primary_subject",
      "notes": "Pose reference"
    },
    {
      "role": "style_reference",
      "notes": "Cartoon style reference"
    }
  ],
  "style": "anime_style",
  "controlnet": {
    "enabled": true,
    "type": "pose",
    "weight": 1.2
  },
  "ip_adapter": {
    "enabled": true,
    "mode": "style",
    "weight": 0.8
  },
  "advanced_params": {
    "denoising_strength": 0.7,
    "guidance_scale": 8.0
  }
}

---

📚 Knowledge Base

This skill comes with knowledge files serving as catalogs and best practices:

1. ai_image_style_categories.md — Style catalog (151+ styles)
2. prompt_depth_reference.md — Prompt logic and structure
3. vfx_effects_menu.md — Complete VFX menu (50+ effects)
4. realistic_skin_preservation_rules.md — Realistic skin rules
5. identity_consistency_rules.md — Identity preservation rules
6. photorealistic_prompting_research_notes.md — Research notes
7. legacy_system_prompt_reference.md — Legacy workflow reference

---

🎨 Style Catalog (151+ Styles)

Photorealism

• photorealistic, ultra_realistic, raw_realism, dslr_look
• natural_light_realism, street_photography, documentary
• soft_commercial, lifestyle_photography, beauty_commercial
• kodak_portra, fujifilm_superia, cinestill_800t, polaroid

Cinematic

• hollywood_cinematic, teal_and_orange, dark_cinematic
• film_noir, moody_cinematic, golden_hour_cinematic
• suspense_thriller_style, romance_cinematic, sci_fi_cinematic

Illustration & Art

• watercolor, oil_painting, gouache, charcoal, sketch_pencil
• clean_vector, flat_illustration, isometric
• anime_style, manga_style, webtoon_style

Fantasy & Sci-Fi

• medieval_fantasy, cyberpunk_neon, steampunk_industrial
• post_apocalyptic, alien_world, underwater_fantasy

...and many more! See the full catalog in the schema

---

🌐 Platform Support

Platform	text-to-image	image-to-image	inpaint	outpaint
Generic	✅	✅	✅	✅
Stable Diffusion	✅	✅	✅	✅
Midjourney	✅	✅	❌	❌
DALL-E 3	✅	✅	✅	⚠️
Gemini/Imagen	✅	✅	✅	✅
Flux	✅	✅	⚠️	❌
Firefly	✅	✅	✅	⚠️

---

⚙️ Task Types

• final_prompt — Generate the final ready-to-use prompt
• background_10 — Generate 10 random background/scene ideas
• ideas_10 — Generate 10 concept ideas
• angles_10 — Generate 10 camera angles/compositions
• storyboard_continue — Continue from existing storyboard frames (adds next frames)
• storyboard_6 — Create a 6-scene storyboard
• infographic_layout — Infographic layout structure
• style_catalog — Show style menu
• vfx_catalog — Show VFX menu
• typography_catalog — Show Typography menu
• update_preferences — Update preferences

---

🔧 Default Values Summary

All inputs have the following default values:

{
  "request": "",                    // required field
  "generation_mode": "text_to_image",
  "background_type": "normal",      // normal | green_screen | blue_screen | transparent
  "task": "final_prompt",
  "detail_level": "standard",
  "languages": "en",
  "aspect_ratio": "9:16",
  "aspect_ratio_custom": "",
  "style": "photorealistic",
  "target_platform": "generic",
  "text_on_image": false,
  "realistic_skin": false,
  "identity_lock": "none",
  "reference_images": [],
  "edit_mask": {},
  "outpaint_config": {},
  "advanced_params": {
    "denoising_strength": 0.75,
    "guidance_scale": 7.5,
    "steps": 50,
    "seed": -1,
    "sampler": "dpm_2m_karras",
    "clip_skip": 1
  },
  "controlnet": {
    "enabled": false,
    "type": "canny",
    "weight": 1.0
  },
  "ip_adapter": {
    "enabled": false,
    "mode": "style",
    "weight": 0.6
  }
}

---

📝 Version History

v2.1 (Current)

• ✅ Added generation_mode for all image generation modes
• ✅ Added text-based inpainting (edit_mask)
• ✅ Added outpainting support (outpaint_config)
• ✅ Added advanced_params (strength, CFG, steps, seed, sampler)
• ✅ Added ControlNet and IP-Adapter support
• ✅ Added target_platform selection
• ✅ Improved validation and error handling
• ✅ All inputs have default values

v1.0 (Legacy)

• Basic text-to-image
• Style catalog and VFX
• Typography support
• Basic reference images

---

🎯 Best Practices

1. Use generation_mode clearly — Always specify the desired mode
2. Text-based masking — Use natural language to specify edit areas
3. Identity lock — Choose the appropriate level (soft for people, strict for products)
4. Denoising strength — 0.3-0.6 = subtle, 0.7-0.9 = strong transformation
5. Platform-specific — Choose the target_platform matching your actual platform

---

🎨 Style Value to Description Mapping

CRITICAL: When generating prompts, NEVER use the style ID/value directly in the prompt. Instead, always use the descriptive text.

How to Handle Style Input

When the user selects a style like pixar_3d_animated, you MUST convert it to a descriptive phrase:

Style Value	Use This Description in Prompt
pixar_3d_animated	"Pixar-style 3D animation, vibrant colors, expressive characters, smooth rendering"
disney_3d_animated	"Disney-style 3D animation, magical atmosphere, beautiful and charming characters"
dreamworks_3d	"DreamWorks-style 3D animation, fun and humorous, dynamic expressions"
claymation	"Claymation stop-motion style, clay-like texture, handcrafted appearance"
low_poly_3d	"Low poly 3D style, geometric shapes, minimalist polygonal aesthetic"
photorealistic	"Photorealistic, ultra-detailed, lifelike textures, natural lighting"
ultra_realistic	"Ultra-realistic, hyper-detailed, 8K resolution, cinematic quality"
ghibli_style_mood	"Studio Ghibli-style, warm and nostalgic, hand-painted aesthetic, whimsical atmosphere"
cyberpunk	"Cyberpunk aesthetic, neon lights, dark urban environment, futuristic technology"
watercolor	"Watercolor painting style, soft color bleeding, translucent washes, artistic texture"
oil_painting	"Oil painting style, rich colors, visible brushstrokes, classical artistic technique"
modern_anime	"Modern anime style, clean lines, vibrant colors, expressive eyes"

General Rule

For any style value:

1. Extract the readable parts from the value (replace underscores with spaces)
2. Add descriptive modifiers that enhance the style
3. NEVER output the raw value like "style: pixar_3d_animated" in the prompt

❌ WRONG: "A cat, style pixar_3d_animated" ✅ CORRECT: "A cat in Pixar-style 3D animation, vibrant colors, expressive features, smooth CGI rendering"

---

📊 Prompt Count and Layout Formats

When prompt_count is specified with a layout format, generate prompts accordingly:

Value	Count	Layout Type	Description
1	1	Single	Single standalone image
2_distinct	2	Distinct	2 separate, unique shots
2_1x2	2	Grid	Vertical arrangement (1 column, 2 rows)
2_2x1	2	Grid	Horizontal arrangement (2 columns, 1 row)
3_distinct	3	Distinct	3 separate, unique shots
3_1x3	3	Grid	Vertical strip
3_3x1	3	Grid	Horizontal strip
4_distinct	4	Distinct	4 separate, unique shots
4_2x2	4	Grid	2×2 square grid
6_distinct	6	Distinct	6 separate, unique shots
6_3x2	6	Grid	3 columns × 2 rows
6_2x3	6	Grid	2 columns × 3 rows
6_storyboard	6	Storyboard	Sequential story frames with narrative flow
9_3x3	9	Grid	3×3 square grid
12_4x3	12	Grid	4 columns × 3 rows
12_3x4	12	Grid	3 columns × 4 rows
16_4x4	16	Grid	4×4 square grid

Layout Type Behavior

• Distinct: Each prompt is independent, focusing on variety and unique compositions
• Grid: Prompts should work together as a cohesive set, similar styling and color palette
• Storyboard: Sequential narrative flow, each frame follows the previous logically

---

📞 Support

For questions or additional suggestions, please contact the development team

Version: 2.1 Last Updated: February 4, 2026 License: Proprietary