Image Architect Agent

The Image Architect is a specialized agent that handles complex image tasks. It understands image requirements, selects optimal generators, and plans efficient multi-step workflows.

Note: The Image Architect uses MCP tools for session state and iteration. Restart Claude Code once after installing the plugin to enable full agent capabilities.

When to Use the Agent

The Image Architect automatically activates for:

• Complex image tasks requiring multiple steps
• Requests that need generator selection advice
• Multi-image workflows (dashboards, asset sets)
• Tasks requiring optimization decisions

Capabilities

Generator Expertise

The agent deeply understands each floimg generator:

Generator	Best For	Key Parameters
OpenAI/DALL-E	Photorealistic images, illustrations, creative scenes	prompt, size, quality
QuickChart	Data visualization, charts, graphs	type, data, options
Mermaid	Technical diagrams, flowcharts, sequences	code
QR	QR codes, barcodes	text, errorCorrectionLevel
Screenshot	Webpage captures	url, fullPage, width
D3	Custom data visualizations	render, data

Generator Selection

The agent analyzes your request and picks the right tool:

AI Images (OpenAI) - Creative, photorealistic, or artistic content

• “a sunset over mountains”
• “product mockup on marble table”
• “illustration of a robot”

Charts (QuickChart) - Data visualization

• “bar chart of sales by quarter”
• “pie chart showing market share”
• “line graph of user growth”

Diagrams (Mermaid) - Technical/architectural visuals

• “flowchart of user registration”
• “sequence diagram of API calls”
• “entity relationship diagram”

QR Codes - Encoded data

• “QR code for website URL”
• “QR with WiFi credentials”

Screenshots (Playwright) - Webpage captures

• “screenshot of competitor’s landing page”
• “capture the mobile view of our site”

Workflow Planning

For complex requests, the agent designs optimized pipelines:

1. Analyze requirements - What final output is needed?
2. Decompose into steps - Generate -> Transform(s) -> Save
3. Choose optimal generators - Match capability to need
4. Plan transforms - Apply in optimal order (resize last for quality)
5. Execute efficiently - Use pipelines for multi-step work

Transform Knowledge

The agent knows when and how to apply transforms:

Operation	When to Use	Quality Tips
resize	Final sizing	Apply last to preserve quality
blur	Privacy, backgrounds	Low sigma (1-3) for subtle
sharpen	After resize	Low sigma (0.5-1)
addCaption	Branding, context	Use contrast colors
roundCorners	UI elements, avatars	Match design system
preset	Quick styling	vintage, vibrant, dramatic, soft

Example Interactions

Dashboard Creation

You: “Create a dashboard with 3 charts showing our quarterly data”

Agent approach:

1. Create three separate chart generations
2. For each: generate with appropriate chart type
3. Optionally resize all to consistent dimensions
4. Report imageIds and paths for each

Social Media Assets

You: “Generate a hero image for our landing page and prepare social versions”

Agent approach:

1. Generate high-quality AI image (1792x1024, hd quality)
2. Create pipeline with resize variants:
   • 1200x630 for Open Graph
   • 800x418 for Twitter
   • 1080x1080 for Instagram
3. Save each variant to specified destination

Branded Documentation

You: “Create a technical architecture diagram and add our company watermark”

Agent approach:

1. Generate Mermaid diagram with proper code
2. Transform: addText with company name/logo position
3. Save to cloud for documentation

Agent Behavior

The Image Architect focuses on getting the job done well:

• Asks clarifying questions only when truly needed
• Explains its generator choices briefly
• Reports results with file paths and imageIds
• Offers follow-up options for variations or transforms

Clarifying Questions

The agent may ask about:

• Final use case - Social media? Documentation? Print?
• Size/format requirements - Dimensions? File format?
• Storage destination - Local? S3? R2?

Result Reporting

After completing a task, the agent reports:

• What was created
• Where it’s saved (path or URL)
• ImageId for follow-up operations
• Options for additional transforms

Triggering the Agent

The agent activates automatically for complex tasks. You can also invoke it directly by describing complex image needs:

I need to create a data visualization dashboard with:
- A bar chart of monthly sales
- A pie chart of product categories
- All charts should be 800x600 and saved to ./charts/

The agent will plan and execute the entire workflow.