@dfeirstein/image-server/docs/prompt-recipes.md

MCP server for creating, editing, and combining images with OpenAI models

# Creative Prompt Recipes for MCP Image Creator

> *A curated collection of prompt recipes for generating exceptional images with the MCP Image Creator*

## Table of Contents

- [Quick-Start Template](#quick-start-template)
- [Image Editing with Reference Images](#image-editing-with-reference-images)
- [Style Definition JSON](#style-definition-json)
- [Prompt Recipes](#prompt-recipes)
  - [Hero Backgrounds](#hero-backgrounds)
  - [Icons](#icons-1024×1024-transparent)
  - [Section Illustrations](#section-illustrations-blog--feature)
  - [Textures & Patterns](#textures--patterns)
  - [Fun / Easter Eggs](#fun--easter-eggs)
- [Photographic Prompt Recipes](#photographic-prompt-recipes)
  - [Hero Banner Photos](#hero-banner-photos)
  - [Feature/Benefit Photos](#feature--benefit-illustrative-photos)
  - [Avatar/Testimonial Portraits](#avatar--testimonial-portraits-square)
  - [Blog/Case-Study Headers](#blog--case-study-header-photos)
- [Tips & Knobs](#tips--knobs)
  - [For Abstract/Design Prompts](#tips--knobs)
  - [For Photographic Prompts](#tips-for-photographic-prompts)

## Quick-Start Template

```jsonc
// For generating images from text prompt:
{
  "name": "create_image",
  "arguments": {
    "prompt": "<your vivid description>",
    "brandSignature": "palette:#0EA5E9,#0F172A; tone:Clean",
    "size": "1536x1024",
    "quality": "high",
    "filename": "<descriptive-name>.png",
    "outputPath": "<subfolder>",
    "targetProjectDir": "/absolute/path/to/project"
  }
}
```

## Image Editing with Reference Images

You can use reference images to edit or combine existing images using OpenAI's `gpt-image-1` model. Add the `referenceImagePaths` parameter with an array of paths to existing images (relative to your target project's public folder).

```jsonc
// For editing/combining existing images:
{
  "name": "create_image",
  "arguments": {
    "prompt": "Create a lovely gift basket with these products inside it",
    "referenceImagePaths": [
      "products/soap.png",
      "products/lotion.png",
      "products/incense.png"
    ],
    "filename": "gift-basket.png",
    "outputPath": "marketing/promo",
    "targetProjectDir": "/absolute/path/to/project"
  }
}
```

**Note**: When using `referenceImagePaths`:
- The model is automatically set to `gpt-image-1` (required for image editing)
- Size parameter is ignored as it's not supported for image editing operations
- All reference images should exist in your target project's public folder (or subdirectory)
- Use detailed prompts describing the desired modifications or combinations

## Style Definition JSON

The `styleDefinitionJSON` parameter allows you to define a comprehensive visual style once and reuse it across multiple image generations. The JSON structure is passed directly to the image generation model, which can interpret the structured data within the prompt text.

This approach is particularly valuable for:

- Creating themed image collections with unified visual language
- Maintaining brand consistency across multiple assets
- Efficiently generating image series where only the subject changes
- Ensuring consistent style elements like lighting, composition, and color palette

### Example Style Definition JSON

```jsonc
{
  "name": "create_image",
  "arguments": {
    "prompt": "Abstract representation of Litigation",
    "styleDefinitionJSON": {
      "output_requirements": {
        "color_mode": "Full Vibrant Color Photography", 
        "style": "Abstract Conceptual Realism",
        "quality": "High-end Professional Photo Shoot"
      },
      "color_palette": {
        "primary_accent": {"name": "Primary Blue", "hex": "#1474F3"},
        "supporting_colors": [
          {"name": "Deep Navy", "hex": "#1C2B4A"},
          {"name": "Warm Neutral", "hex": "#E8D7C1"}
        ],
        "overall_tone": "Predominantly cool with warm accents"
      },
      "visual_elements": {
        "lighting": "Soft directional studio light emphasizing color and texture",
        "composition": "Clean, minimalist, shallow depth of field",
        "textures": ["Polished metal", "Premium textured paper", "Smooth glass"],
        "perspective": "Professional product/still life angle"
      },
      "strict_exclusions": [
        "No Black and White or Monochrome output",
        "No Text or Writing",
        "No People or Figures"
      ]
    },
    "quality": "high",
    "size": "1024x1024",
    "filename": "litigation-concept.png",
    "outputPath": "concepts",
    "targetProjectDir": "/absolute/path/to/project"
  }
}
```

### Style Definition Structure

Your `styleDefinitionJSON` can use any structure that effectively communicates your style requirements. The image model receives the raw JSON structure within the prompt text and can interpret its meaning.

Some effective patterns include:

1. **`output_requirements`**: Overall output specifications
2. **`color_palette`**: Color scheme definitions including hex values
3. **`visual_elements`**: Physical aspects like lighting, composition, and textures
4. **`strict_exclusions`**: Explicit list of elements to avoid in the image

The JSON structure is flexible, allowing you to organize style information in whatever way makes most sense for your use case.

### Prompt Construction

When both `prompt` and `styleDefinitionJSON` are provided, the system constructs a prompt for the image model in this format:

```
[User Prompt]

--- STYLE DEFINITION (JSON) ---
{
  "detailed": "style definition",
  "in": "JSON format",
  "with": "all properties"
}
```

This allows the image model to receive both your textual description and the structured style definition, using both to guide the image generation process.

## Prompt Recipes

### Hero Backgrounds
| Goal | Prompt (copy-paste ready) |
|------|--------------------------|
| Abstract gradient mesh | `"Dreamy gradient mesh swirling between #0EA5E9 and #0F172A, subtle bokeh lights, modern SaaS hero backdrop, soft focus, ultra-wide composition"` |
| Futuristic skyline silhouette | `"Minimalist cyber skyline at dusk, neon line art, horizon glow, negative space left for headline, high resolution matte painting"` |
| Paper-cut waves | `"Layered paper-cut ocean waves, brand blues, slight shadow depth, clean vector aesthetic, high contrast for overlay text"` |

### Icons (1024×1024, transparent)
| Purpose | Prompt |
|---------|--------|
| Export icon | `"Line icon, arrow exiting square, 2.5pt stroke, rounded ends, brand color #0EA5E9 on transparent, flat SVG style"` |
| AI chip | `"Isometric microchip with glowing neural lines, duotone brand palette, crisp pixel-perfect edges"` |

### Section Illustrations (blog / feature)
| Scenario | Prompt |
|----------|--------|
| Flow state coder | `"Stylized illustration of developer at standing desk, swirling code snippets, teal and midnight palette, open source vibes, flat shading"` |
| A/B variant cards | `"Two overlapping website cards labeled A and B, playful pastel colors, subtle drop shadow, 3:2 aspect"` |

### Textures & Patterns
| Use | Prompt |
|-----|--------|
| Subtle grid overlay | `"Ultra-light blueprint grid pattern, 8px spacing, brand teal lines on transparent, repeating seam-less tile"` |
| Particle network | `"Constellation of data nodes linked by thin glowing lines, dark background, depth-of-field blur, 4K wallpaper"` |

### Fun / Easter Eggs
| Idea | Prompt |
|------|--------|
| Mascot otter | `"Cute otter mascot wearing VR goggles, holding paint brush, tech startup style, vector mascot, brand palette accent"` |
| 404 illustration | `"Robot inspecting broken cable, large 404 numbers, humorous, monochrome line art with teal highlight"` |

## Photographic Prompt Recipes

### Hero Banner Photos  
| Desired Feel | Prompt |
|--------------|--------|
| Tech-futuristic office | `"Wide-angle photograph of a modern glass office atrium at sunset, teal & navy brand lighting accents, shallow depth of field foreground plants, cinematic look, 16:9 hero banner"` |
| Creative flow | `"Over-the-shoulder shot of a developer at ultra-wide monitor, dark mode code glowing, soft rim light, brand palette LED strip, candid studio photography, 1536×1024"` |
| Startup teamwork | `"Candid photo of diverse 3-person dev team whiteboarding UX wireframes, high-key lighting, authentic smiles, neutral backdrop, natural color grading"` |

### Feature / Benefit Illustrative Photos  
| Feature Angle | Prompt |
|---------------|--------|
| Speed & focus | `"Macro shot of mechanical watch gears in motion, dramatic lighting, blue & cyan gels, sharp detail, conveys precision speed"` |
| Brand consistency | `"Flat-lay of Pantone color swatches, teal / navy family, designer desk scene, soft daylight, minimal clutter, top-down photograph"` |
| Zero context-switch | `"Side profile of developer wearing noise-canceling headphones, blurred background monitors, subtle bokeh, calm color palette"` |

### Avatar / Testimonial Portraits (square)  
| Persona | Prompt |
|---------|--------|
| Happy SaaS founder | `"Natural light headshot of 30-yo Latina founder in modern loft office, genuine smile, shallow depth of field, 1:1 aspect, brand teal accent scarf"` |
| Senior developer | `"Studio portrait of bearded software engineer, soft gray backdrop, moody Rembrandt lighting, confident expression, square crop"` |
| Customer success manager | `"Bright portrait of friendly customer success rep wearing headset, clean white background, even softbox lighting, 1024×1024"` |

### Blog / Case-Study Header Photos  
| Story Angle | Prompt |
|-------------|--------|
| A/B test results | `"Overhead shot of two laptop screens side by side, variant A vs B dashboards, warm café lighting, storytelling composition"` |
| AI workflow | `"Close-up of code editor reflecting in developer's glasses, dark environment, neon teal reflections, cinematic shallow DOF"` |

## Tips & Knobs

*  **Brand tuning:** keep a palette list in `branding/palette.json`; reference via `brandSignature`.
*  **Model swap:** heavier creativity → `gpt-image-1`; faster/cheaper → `dall-e-2`.
*  **Variant batching:** loop through prompts with filenames `hero-v1.png`, `hero-v2.png` for split tests.
*  **Negative cues:** add phrases like `"no text, no watermark"` to reduce noise.
*  **Aspect hints:** make size match use case: hero = 1536×1024, icon = 1024×1024, tall illustration = 1024×1536.

## Tips for Photographic Prompts  
1. **Lighting adjectives**: *softbox, rim light, golden-hour, neon gels* to steer mood.  
2. **Lens / depth cues**: *macro*, *bokeh*, *wide-angle* for composition.  
3. **Authenticity**: add *candid*, *natural light*, *genuine* to reduce stock-photo vibe.  
4. **Brand tint**: reference palette colors subtly (*teal accents*, *navy background*).  
5. **People diversity**: call out ethnicity/age range when representation matters.  
6. **Aspect ratio**: hero = 16:9 or 1536×1024, avatars = 1:1.