Style Library¶

Styles are reusable presets that capture a visual direction — palette, composition, mood, medium, and constraints — for consistent image generation across conversations.

How Styles Work¶

A style is a creative brief, not a prompt template. When you apply a style, the LLM reads the brief and adapts it to the target provider's prompt format:

OpenAI (gpt-image-1.5 / gpt-image-1 / gpt-image-1-mini) — natural language description incorporating the style's direction.
Gemini (gemini-2.5-flash-image / gemini-3.1-flash-image-preview / gemini-3-pro-image-preview) — natural language, similar to OpenAI; can also incorporate the style brief into multi-image compositing or conversational refinement.
SD WebUI (SD 1.5 / SDXL / RealVisXL / Juggernaut) — comma-separated CLIP tags with a separate negative prompt.
SD WebUI (Flux 1 / Flux Schnell / FLUX.2) — natural language, similar to OpenAI; no native negative prompts (CFG=1 distilled).
SD WebUI (SD 3 / 3.5) — natural language, similar to Flux, but with native negative prompts (unlike Flux). Skip SDXL-style (weight:1.2) parens.
SD WebUI (Pony Diffusion XL / Illustrious-XL / NoobAI-XL) — Booru-style tag grammar with a mandatory score_* prefix on Pony.

The style text is never copied verbatim into the generation prompt — the LLM rewrites it in the right grammar for the chosen model.

The per-model registry (list_providers → style_profile.style_hints / incompatible_styles) is the canonical source for what each model wants; cross-reference it from your style brief when you need provider-specific guidance.

style_profile vs style library. The two are deliberately separate concerns:

Style library (this guide) — a user-saved creative brief applied to a generation request. Describes the brief.

style_profile (set by the registry, surfaced via list_providers) — per-model metadata describing what each model is best at. Describes the model.

Both can use the word "style" in their everyday English sense — they don't interact. See ADR-0009 for the disambiguation.

Style File Format¶

Each style is a Markdown file with YAML frontmatter stored in the styles directory (default: ~/.image-generation-mcp/styles/).

---
name: website
tags: [brand, web, modern]
provider: auto
aspect_ratio: "16:9"
quality: hd
---

Minimalist flat illustration. Geometric shapes, clean lines.
Brand palette: deep teal (#0D4F4F), warm cream (#F5F0E8), coral accent (#FF6B5E).
Plenty of negative space. No photorealism, no gradients, no text in image.
Suitable for hero banners and section dividers.

Frontmatter Fields¶

Field	Type	Required	Default	Description
`name`	str	Yes	—	Style identifier (must match filename without `.md`)
`tags`	list	No	`[]`	Categorization tags for browsing/filtering
`provider`	str	No	`auto`	Suggested provider (`auto`, `openai`, `sd_webui`, etc.)
`aspect_ratio`	str	No	—	Default aspect ratio (e.g. `16:9`, `1:1`)
`quality`	str	No	—	Default quality level (`standard` or `hd`)

Body¶

The body is free-form Markdown prose describing the visual direction. Write it as you would a creative brief for a designer:

Palette — specific colors with hex codes
Composition — layout rules, negative space, framing
Medium — illustration, photography, watercolor, etc.
Mood — professional, playful, dramatic, etc.
Constraints — what to avoid (photorealism, gradients, text, etc.)

Creating Styles¶

Manually (file on disk)¶

Create a .md file in the styles directory:

# Default location
mkdir -p ~/.image-generation-mcp/styles
cat > ~/.image-generation-mcp/styles/website.md << 'EOF'
---
name: website
tags: [brand, web]
aspect_ratio: "16:9"
quality: hd
---

Minimalist flat illustration with geometric shapes and clean lines.
EOF

Via the `save_style` Tool¶

Save a style from within an MCP conversation:

User: Save this as my "social-media" style

Tool call: save_style
  name: "social-media"
  body: "Vibrant, eye-catching photography style. Bold saturated colors,
         high contrast. Square framing optimized for Instagram and Twitter.
         Clean backgrounds, single focal point."
  tags: ["social", "photography"]
  aspect_ratio: "1:1"
  quality: "hd"

Browsing Styles¶

Resources¶

style://list — JSON array of all styles with names, tags, and descriptions
style://{name} — full Markdown content of a specific style

Example¶

Tool call: read_resource
  uri: "style://list"

→ [
    {
      "name": "website",
      "tags": ["brand", "web", "modern"],
      "description": "Minimalist flat illustration. Geometric shapes, clean lines.",
      "provider": "auto",
      "aspect_ratio": "16:9",
      "quality": "hd"
    }
  ]

Applying Styles¶

Use the apply_style prompt to generate an image with a style:

User: Create a hero banner for the about page using the website style

Prompt: apply_style
  style_name: "website"
  user_request: "hero banner for the about page"

→ Claude reads the style brief, then calls generate_image with a
  provider-appropriate prompt incorporating the style's visual direction

Example Conversation Flow¶

User: Show me what styles are available

Claude: [reads style://list]
  You have 3 styles saved:
  - website (brand, web) — minimalist flat illustrations
  - social-media (social) — vibrant photography
  - presentation (business) — clean corporate graphics

User: Generate a hero image for our pricing page using the website style

Claude: [loads apply_style prompt with style_name="website"]
  [reads the creative brief: "Minimalist flat illustration..."]
  [composes provider-appropriate prompt based on the brief]

Tool call: generate_image
  prompt: "Minimalist flat illustration of a pricing comparison layout,
           geometric shapes representing pricing tiers, deep teal and
           warm cream palette with coral accent highlights, clean lines,
           plenty of negative space, no text"
  aspect_ratio: "16:9"
  quality: "hd"

Managing Styles¶

Deleting a Style¶

Tool call: delete_style
  name: "old-style"

→ {"name": "old-style", "deleted": true}

Updating a Style¶

Call save_style with the same name to overwrite:

Tool call: save_style
  name: "website"
  body: "Updated creative brief with new brand colors..."
  tags: ["brand", "web", "2024"]

→ {"name": "website", "created": false}

Configuration¶

Variable	Type	Default	Description
`IMAGE_GENERATION_MCP_STYLES_DIR`	Path	`~/.image-generation-mcp/styles/`	Directory for style preset files

The directory is created automatically if it does not exist.