image-gen（图像生成器）

Image Gen

Contents

• Overview
• When to Use This Skill
• Workflow

- Step 1: Analyze Article Content - Step 2: Generate Cover Image - Step 3: Generate In-Article Images - Step 4a: Integrate Images - Step 4b: Write ALT Text - Step 5: Verify and Save

• Command Reference
• Prompt Engineering Best Practices
• Common Patterns
• Resources

Overview

Generate high-quality images for technical articles using the imagen CLI tool (powered by gemini-2.5-flash-image). This skill provides a systematic workflow for creating two types of images:

1. Cover Images: High-impact, conceptual images that grab attention on LinkedIn/Medium
2. In-Article Images: Technical diagrams and illustrations that explain specific concepts

The skill includes comprehensive prompt engineering guidance and best practices for integrating images with SEO-friendly ALT text.

When to Use This Skill

Invoke this skill when:

• Creating visual assets for a completed or near-complete article
• User requests "generate images for this article" or similar
• Article needs a cover image for publication
• Technical concepts would benefit from visual explanation
• Breaking up long text sections with relevant imagery
• Optimizing article for LinkedIn/Medium with visual appeal

Typical workflow position: After article content is finalized (Phase 8 in the article enhancement workflow), before SEO optimization.

Workflow

Step 1: Analyze Article Content

Before generating any images, analyze the article to identify:

1. Article topic and key themes: What is the main subject?
2. Key concepts needing visualization: Which technical ideas are complex or abstract?
3. Section breaks: Where would images improve readability?
4. Target audience: Developers? DevOps? Beginners or advanced?

Output: List of 2-4 images needed (1 cover + 1-3 in-article)

Example:

Article: "Mastering Alembic for Database Migrations"
Images needed:
- Cover: Conceptual image representing database evolution/migration
- Image 1: Diagram showing migration workflow
- Image 2: Visual comparing manual SQL vs Alembic automation

Step 2: Generate Cover Image

Cover images should be high-impact, conceptual, and visually interesting to grab attention. They are often metaphorical rather than literal.

Prompt Engineering for Covers:

• Style keywords: digital art, 3D render, conceptual art, abstract
• Atmosphere: cinematic lighting, dramatic, futuristic, professional
• Composition: Specify angle, focus, and elements
• Structure: [Style] cover image of [Metaphor/Concept] representing [Article Topic], [Atmosphere/Lighting], [Composition]

Command:

imagen generate "DETAILED PROMPT" --output work/images/[article_name]_cover.png

Example Prompts:

For a Vertex AI article:

imagen generate "A stunning digital art cover image showing three glowing pathways converging into a central AI cloud platform. Each pathway is a different color (green, blue, gold) representing different AI models flowing into Google Cloud infrastructure. The scene has dramatic cinematic lighting with a futuristic tech aesthetic. Style: conceptual, 3D render, high-detail, professional." --output work/images/vertex_ai_cover.png

For an Alembic article:

imagen generate "A dramatic 3D render showing a database schema transforming like evolving architecture, with migration files as glowing blueprints guiding the transformation. Dark background with cinematic lighting highlighting the evolution process. Style: conceptual art, digital art, high-detail, professional." --output work/images/alembic_cover.png

Tips:

• Be specific and detailed (100+ words is fine)
• Use metaphors that resonate with the topic
• Specify lighting/atmosphere for mood
• Include "professional" or "high-detail" for quality

Step 3: Generate In-Article Images (1-3)

In-article images should explain or reinforce specific concepts from the text. They are more direct and educational than cover images.

Prompt Engineering for In-Article Images:

• Style keywords: technical diagram, blueprint, minimalist illustration, clean
• Clarity: clear labels, connecting arrows, color-coded, organized
• Purpose: Focus on explaining one specific concept
• Structure: [Style] showing [Specific Concept], [Visual Elements], [Labels/Clarity]

Command:

imagen generate "CONCEPT-SPECIFIC PROMPT" --output work/images/[article_name]_[concept].png

Example Prompts:

For a technical diagram:

imagen generate "A clean technical diagram showing three distinct client libraries connecting to a central Vertex AI platform. Use color-coded arrows: green for native Google SDK, blue for partner Anthropic SDK, and gold for OpenAI-compatible SDK. Include clear labels for each connection path. Style: blueprint, technical diagram, clear labels, professional." --output work/images/vertex_ai_three_patterns.png

For a comparison illustration:

imagen generate "A minimalist side-by-side comparison showing traditional infrastructure (left) with tangled wires and manual configuration versus modern infrastructure-as-code (right) with clean, organized flow. Use contrasting colors and simple icons. Style: clean, minimalist illustration, symbolic." --output work/images/traditional_vs_iac.png

For a workflow diagram:

imagen generate "A flowchart-style diagram showing the Alembic migration workflow: 1) Model changes → 2) Auto-generate migration → 3) Review migration → 4) Apply to database. Use arrows connecting each step with clear labels. Style: blueprint, technical flowchart, organized, clean." --output work/images/alembic_workflow.png

Tips:

• Focus on one concept per image
• Use color-coding to distinguish elements
• Request "clear labels" or "connecting arrows" for clarity
• Simpler is often better for technical diagrams

Step 4a: Integrate Images

CRITICAL: Integrate generated images into the article at strategic locations.

Integration Checklist:

1. Create work/images/ directory if needed
2. Verify all images generated successfully (cover + in-article)
3. Add cover image immediately after H1 title
4. Add in-article images at strategic section breaks
5. Save updated article as new version (e.g., v8_with_images.md)

Markdown Integration Format:

# Article Title

![Cover image description here](images/article_cover.png)

## Introduction

Text introducing the problem...

![Diagram description here](images/article_diagram.png)

Step 4b: Write ALT Text

Write descriptive ALT text (50-125 characters) for each image to ensure accessibility and SEO.

Good ALT Text (descriptive, specific):

• "Architecture diagram showing Cloud Run connecting through VPC Connector to AlloyDB in private subnet"
• "Comparison showing manual infrastructure with tangled wires versus automated infrastructure-as-code"
• "Flowchart illustrating Alembic migration workflow from model changes to database update"

Bad ALT Text (vague, non-descriptive):

• "diagram" or "image"
• "architecture" (too vague)
• "image1.png" or "screenshot"

Step 5: Verify and Save

Final checks:

1. All images generated successfully (check file paths)
2. Images linked correctly in markdown
3. ALT text is descriptive and SEO-friendly (50-125 chars)
4. Cover image placed after H1 title
5. In-article images at logical section breaks
6. New version saved (e.g., work/draft/article_v8_with_images.md)

Typical outputs:

• 1 cover image (*_cover.png)
• 1-3 in-article images (*_concept.png, *_workflow.png, etc.)
• Updated article with integrated images and ALT text

Command Reference

Basic Usage:

imagen generate "PROMPT" --output FILENAME.png

Directory Setup:

mkdir -p work/images

Batch Generation Example:

# Cover
imagen generate "COVER PROMPT" --output work/images/article_cover.png

# In-article images
imagen generate "DIAGRAM PROMPT" --output work/images/article_diagram.png
imagen generate "COMPARISON PROMPT" --output work/images/article_comparison.png

Prompt Engineering Best Practices

For comprehensive prompt engineering guidance, refer to references/prompting_guide.md.

Core Principles:

1. Be Specific: Detailed descriptions produce better results
2. Define Style: Always specify artistic style (digital art, blueprint, etc.)
3. Set Atmosphere: Use lighting and mood keywords
4. Use Photography Terms: wide-angle, close-up, depth of field, etc.

Cover vs In-Article Summary:

Common Patterns

Pattern 1: Simple Article (1-2 images)

• Cover image only
• OR Cover + 1 key diagram

Pattern 2: Standard Article (2-3 images)

• Cover image (conceptual)
• 1-2 in-article images (technical diagrams)

Pattern 3: Complex Tutorial (3-4 images)

• Cover image (conceptual)
• Workflow/architecture diagram
• Comparison illustration
• Example/output visualization

Resources

references/prompting_guide.md

Complete guide to prompt engineering for the imagen CLI, including:

• Detailed prompt crafting principles
• Style and medium keywords
• Lighting and atmosphere techniques
• Photography/cinematography terms
• Extensive examples for different image types

Load this reference when crafting complex prompts or needing inspiration for visual concepts.