image-generation（图像生成）

Image Generation Skill

Overview

Image Generation uses FAL.ai nanobanana pro to create professional-quality images from text descriptions. This skill teaches you how to craft effective prompts and use the automation system to generate assets directly in Claude Code.

Keywords: image generation, nanobanana pro, prompt engineering, AI art, visual content creation, asset generation, automation

Core Models

nanobanana pro — Recommended

• Quality: Highest, most detailed
• Speed: 30-60 seconds
• Best For: Product photography, hero images, final assets
• Use Case: When quality matters most

nanobanana pro

• Quality: High, good detail
• Speed: 5-15 seconds
• Best For: Testing, iterations, social media
• Use Case: When speed matters

nanobanana pro

• Quality: Latest, improved
• Speed: 20-40 seconds
• Best For: Production work
• Use Case: When you want the latest model

Prompt Engineering Framework

The 5-Part Prompt Formula

Every effective prompt has 5 components:

1. Subject — What is the main thing?

"A luxury leather watch"
"A modern logo"
"An Instagram post graphic"

2. Description — What does it look like?

"with gold accents and brown leather strap"
"geometric style, minimalist design"
"vibrant colors, eye-catching composition"

3. Style — What's the artistic style?

"professional product photography"
"modern illustration"
"digital design"
"photorealistic"

4. Technical Details — Quality and format specs

"studio lighting, sharp focus, 4K, centered composition"
"high contrast, trending design, professional quality"
"detailed, well-lit, professional photography"

5. Mood/Aesthetic — What's the feeling?

"luxury and professional"
"energetic and modern"
"clean and minimalist"
"warm and inviting"

Complete Prompt Example

A luxury leather watch with gold accents and brown strap,
professional product photography, studio lighting with rim light,
centered composition, sharp focus, 4K, luxury and professional mood

Prompt Engineering Techniques

Technique 1: Be Specific

❌ Bad: "A watch"
✅ Good: "A luxury leather watch with gold accents on white background"

Technique 2: Use Descriptive Adjectives

❌ Bad: "A logo"
✅ Good: "A modern, geometric, minimalist logo in blue and white"

Technique 3: Reference Styles

❌ Bad: "A nice graphic"
✅ Good: "A graphic in the style of modern Instagram design trends"

Technique 4: Specify Quality

❌ Bad: "A photo"
✅ Good: "A professional 4K product photograph with studio lighting"

Technique 5: Include Composition

❌ Bad: "A person"
✅ Good: "A person in rule of thirds composition, natural lighting, centered"

Claude Code Integration

How to Use in Claude Code

Claude Code can directly generate images using the automation system:

from claude_integration import generate_asset

# Generate a single image
result = generate_asset(
    category="product-photos",
    name="luxury-watch",
    prompt="A luxury leather watch with gold accents on white background, professional product photography, studio lighting, 4K, sharp focus",
    size="1024x1024",
    num_variations=1
)

print(f"Generated: {result['images']}")

Setup for Claude Code

Ensure these files are in your project:

your-project/
├── vibe-creative-automation/
│   ├── fal_api.py
│   ├── creative_cli.py
│   ├── claude_integration.py
│   └── requirements.txt
└── assets/  (will be created automatically)

Set environment variable:

export FAL_API_KEY="your_key_here"

Install dependencies:

pip install requests

Claude Code Workflow

When you ask Claude: "Generate 3 variations of a product photo for my watch"

Claude will:

1. Read Image Generation skill to understand prompting
2. Read Creative Strategist to get your style
3. Craft the prompt combining both
4. Call automation system with the prompt
5. Generate 3 images using FLUX model
6. Save to folder like assets/product-photos/luxury-watch/
7. Show you results with file paths

Example: Claude Code Generates Product Photos

from claude_integration import generate_asset

# Your Creative Strategist style (from your style guide)
YOUR_STYLE = {
    "primary_style": "photorealistic",
    "mood": "professional and luxurious",
    "lighting": "studio lighting with rim light",
    "composition": "centered"
}

# Generate product photo using your style
prompt = f"""
A luxury leather watch with gold accents,
{YOUR_STYLE['primary_style']},
{YOUR_STYLE['mood']},
{YOUR_STYLE['lighting']},
{YOUR_STYLE['composition']},
white background,
4K,
sharp focus,
professional product photography
"""

result = generate_asset(
    category="product-photography",
    name="luxury-watch",
    prompt=prompt,
    size="1024x1024",
    num_variations=3
)

for img_path in result['images']:
    print(f"✅ Generated: {img_path}")

Batch Generation Example

from claude_integration import batch_generate_assets

# Generate multiple assets at once
assets = [
    {
        "type": "custom",
        "category": "product-photos",
        "name": "watch",
        "prompt": "Luxury watch, professional photography, studio lighting, 4K"
    },
    {
        "type": "custom",
        "category": "product-photos",
        "name": "wallet",
        "prompt": "Premium leather wallet, professional photography, studio lighting, 4K"
    },
    {
        "type": "custom",
        "category": "product-photos",
        "name": "sunglasses",
        "prompt": "Designer sunglasses, professional photography, studio lighting, 4K"
    }
]

results = batch_generate_assets(assets)

for result in results:
    print(f"{result['asset_name']}: {result['images']}")

Image Sizes

Choose the right size for your use case:

Generation Parameters

Guidance Scale (3.5 - 7.5)

Controls how strictly the model follows your prompt:

3.5 — More creative freedom, less literal
5.0 — Balanced (recommended)
7.5 — Strict adherence to prompt, more literal

Example:

# More creative
result = generate_asset(..., guidance_scale=3.5)

# Balanced (default)
result = generate_asset(..., guidance_scale=5.0)

# Strict
result = generate_asset(..., guidance_scale=7.5)

Inference Steps (20 - 50)

More steps = higher quality but slower:

20 — Fast, acceptable quality
28 — Balanced (default)
40 — High quality
50 — Maximum quality

Example:

# Fast generation
result = generate_asset(..., inference_steps=20)

# Balanced (default)
result = generate_asset(..., inference_steps=28)

# High quality
result = generate_asset(..., inference_steps=40)

Practical Prompt Examples

Product Photography

A luxury leather watch with gold accents on white background,
professional product photography, studio lighting with rim light,
centered composition, sharp focus, 4K, highly detailed

Social Media Graphic

Instagram post graphic for product launch, vibrant colors,
eye-catching composition, modern design, 1080x1080 format,
trending aesthetic, professional quality

Logo Design

Modern tech company logo, geometric style, blue and white colors,
minimalist design, scalable, professional, clean lines,
suitable for all media

Illustration

Colorful illustration of a person working at a computer,
modern illustration style, bright colors, friendly mood,
professional quality, trending design

Hero Image

A futuristic tech workspace with multiple monitors,
professional photography, modern aesthetic, blue and purple lighting,
cinematic composition, 4K, highly detailed

Integration with Other Skills

Image Generation + Creative Strategist:

• Use your style guide to craft better prompts
• Maintain consistency across all generated images

Image Generation + Product Photography:

• Generate product shots for e-commerce
• Create lifestyle product photos

Image Generation + Social Graphics:

• Generate graphics for social media
• Create platform-specific content

Image Generation + Brand Asset:

• Generate logos and icons
• Create brand illustrations

Command Line Usage

You can also use the CLI directly:

# Generate custom asset
python creative_cli.py custom \
  --category "product-photos" \
  --name "luxury-watch" \
  --prompt "A luxury leather watch with gold accents on white background, professional product photography, studio lighting, 4K, sharp focus" \
  --size 1024x1024 \
  --num-images 3 \
  --model fal-ai/nano-banana-pro

Troubleshooting

Problem: Images don't match my style

Solution:

• Add more specific style descriptors to prompt
• Reference your Creative Strategist guide
• Test with different guidance scales
• Generate multiple variations

Problem: Generation is too slow

Solution:

• Use fal-ai/nano-banana-pro model
• Reduce image size to 768x768
• Reduce inference steps to 20

Problem: Images are too creative/not literal enough

Solution:

• Increase guidance scale to 7.5
• Be more specific in prompt
• Add more technical details

Problem: API errors

Solution:

• Verify FAL_API_KEY is set correctly
• Check internet connection
• Verify API key is valid
• Try again after a moment

Best Practices

1. Start with Creative Strategist — Define your style first
2. Be Specific — More details = better results
3. Test Variations — Generate multiple versions
4. Iterate — Refine based on results
5. Use Consistent Prompts — Similar prompts = consistent style
6. Reference Your Style — Include style descriptors in every prompt
7. Batch Generate — Generate multiple assets at once
8. Organize Assets — Keep generated images organized

Next Steps

1. Define Your Style — Complete Creative Strategist first
2. Craft Your Prompt — Use the 5-part formula
3. Test Generation — Generate a test image
4. Iterate — Refine based on results
5. Batch Generate — Create multiple assets
6. Use in Projects — Integrate with other skills

You now have the power to generate professional images with AI. Start creating! 🎨