SuperAPI AI Directory
Partner中文AI newsFavorites
Skillsimage-generation-mcp

image-generation-mcp

Design & creation

View repository

Updated 10-15 17:25

About this skill

Image Generation via MCP

An AI image generation skill via MCP. Generate high-quality images for marketing, UI, and presentations using Gemini models or compatible services.

When to use this skill

  • Marketing assets: hero images, banners, social media content
  • UI/UX design: placeholder images, icons, illustrations
  • Presentations: slide backgrounds, product visualization
  • Brand consistency: style guide-based image generation

Instructions

Step 1: Configure MCP Environment

# MCP 서버 설정 확인
claude mcp list

# Gemini CLI 사용 가능 여부 확인
# gemini-cli가 설치되어 있어야 함

Required setup:

  • Model name (gemini-2.5-flash, gemini-3-pro 등)
  • API key reference (stored as an environment variable)
  • Output directory

Step 2: Define the Prompt

Writing structured prompts:

**Subject**: [주요 피사체]
**Style**: [스타일 - 미니멀, 일러스트, 사진풍, 3D 등]
**Lighting**: [조명 - 자연광, 스튜디오, 골든아워 등]
**Mood**: [분위기 - 차분한, 역동적, 전문적 등]
**Composition**: [구성 - 중앙 배치, 삼분할 등]
**Aspect Ratio**: [비율 - 16:9, 1:1, 9:16]
**Brand Colors**: [브랜드 컬러 제약사항]

Step 3: Choose the Model

Model Purpose Features
gemini-3-pro-image High quality Complex compositions, detail
gemini-2.5-flash-image Rapid iteration Prototyping, testing
gemini-2.5-pro-image Balanced Quality/speed balance

Step 4: Generate and Review

# 2-4개 변형 생성
ask-gemini "Create a serene mountain landscape at sunset,
  wide 16:9, minimal style, soft gradients in brand blue #2563EB"

# 단일 변수 변경으로 반복
ask-gemini "Same prompt but with warm orange tones"

Review checklist:

  • Brand suitability
  • Composition clarity
  • Aspect ratio accuracy
  • Text readability (when text is included)

Step 5: Deliverables

Final deliverables:

  • Final image files
  • Record prompt metadata
  • Model, aspect ratio, usage notes
{
  "prompt": "serene mountain landscape at sunset...",
  "model": "gemini-3-pro-image",
  "aspect_ratio": "16:9",
  "style": "minimal",
  "brand_colors": ["#2563EB"],
  "output_file": "hero-image-v1.png",
  "timestamp": "2026-01-21T10:30:00Z"
}

Examples

Example 1: Hero Image

Prompt:

Create a serene mountain landscape at sunset,
wide 16:9, minimal style, soft gradients in brand blue #2563EB.
Focus on clean lines and modern aesthetic.

Expected output:

  • 16:9 hero image
  • Prompt parameters saved
  • 2-3 variants for selection

Example 2: Product Thumbnail

Prompt:

Generate a 1:1 thumbnail of a futuristic dashboard UI
with clean interface, soft lighting, and professional feel.
Include subtle glow effects and dark theme.

Expected output:

  • 1:1 square image
  • Low visual noise
  • App store ready

Example 3: Social Media Banner

Prompt:

Create a LinkedIn banner (1584x396) for a SaaS startup.
Modern gradient background with abstract geometric shapes.
Colors: #6366F1 to #8B5CF6.
Leave space for text overlay on the left side.

Expected output:

  • LinkedIn-optimized dimensions
  • Safe zone for text
  • Brand-aligned colors

Best practices

  1. Specify ratio early: prevent unintended cropping
  2. Use style anchors: maintain a consistent aesthetic style
  3. Iterate with constraints: change only one variable at a time
  4. Track prompts: ensure reproducibility
  5. Batch similar requests: generate consistent style sets

Common pitfalls

  • Vague prompts: need to specify a concrete style and composition
  • Ignoring size constraints: check the size requirements of the target channel
  • Overly complex scenes: simplify for clarity

Troubleshooting

Issue: Outputs are inconsistent

Cause: missing stable style constraints Solution: add style references and a fixed palette

Issue: Wrong aspect ratio

Cause: unspecified or unsupported aspect ratio Solution: provide the exact aspect ratio and regenerate

Issue: Brand mismatch

Cause: color code not specified Solution: specify brand colors with HEX codes

Output format

## Image Generation Report

### Request
- **Prompt**: [full prompt]
- **Model**: [model used]
- **Ratio**: [aspect ratio]

### Output Files
1. `filename-v1.png` - [description]
2. `filename-v2.png` - [variant description]

### Metadata
- Generated: [timestamp]
- Iterations: [count]
- Selected: [final choice]

### Usage Notes
[Any notes for implementation]

Multi-Agent Workflow

Validation & Retrospectives

  • Round 1 (Orchestrator): prompt completeness, aspect ratio consistency
  • Round 2 (Analyst): style consistency, brand consistency
  • Round 3 (Executor): verify output filenames and handoff checklist

Agent Roles

Agent Role
Claude Prompt composition, quality verification
Gemini Execute image generation
Codex File management, batch processing

Metadata

Version

  • Current Version: 1.0.0
  • Last Updated: 2026-01-21
  • Compatible Platforms: Claude, ChatGPT, Gemini, Codex

Related Skills

Tags

#image-generation #gemini #mcp #design #creative #ai-art