image_generation

Capability Overview

• Text-to-Image: Generate images based on text descriptions
• Image-to-Image: Generate new images based on reference images
• Image Editing: Modify specific parts of existing images
• Style Transfer: Change image styles (realistic, anime, oil painting, etc.)
• Text Rendering: Generate clear and readable text in images

Workflow

Phase 1: Requirement Understanding

1. Understand the user's image requirements (theme, style, usage)
2. Confirm output format (dimensions, resolution, quantity)
3. If there are reference images, confirm the editing intent

Phase 2: Prompt Construction

1. Convert user intent into English prompts (better results)
2. Follow the prompt formula:

   <subject> <action> <scene> <style> <quality>

3. Supplement necessary detailed descriptions

Phase 3: Image Generation

1. Call the

   generate_image

   tool
2. Call the

   edit_image

   tool if editing is needed
3. Generate multiple candidates (for user selection if required)

Phase 4: Delivery

1. Display the generated results
2. Ask if adjustments are needed
3. Save to the user-specified location

Tool Usage

generate_image

• Purpose: Generate images based on text descriptions
• Parameters:

  • prompt

    : Image description (English yields better results)

  • style

    : Style presets (realistic, anime, oil_painting, watercolor, minimal, cinematic)

  • aspect_ratio

    : Aspect ratio (1:1, 16:9, 9:16, 4:3, 3:4)

  • resolution

    : Resolution (1K, 2K, 4K)

  • num_images

    : Number of images to generate (1-4)
• Example:
  python

  generate_image(
      prompt="A majestic horse galloping through cherry blossoms, golden hour lighting, Chinese New Year festive atmosphere",
      style="realistic",
      aspect_ratio="16:9",
      resolution="2K",
      num_images=2
  )

edit_image

• Purpose: Edit existing images
• Parameters:

  • image_path

    : Original image path or URL

  • prompt

    : Editing instructions (e.g., "Change the background to night scene")

  • preserve_subject

    : Whether to keep the subject unchanged (default True)
• Example:
  python

  edit_image(
      image_path="/workspace/photo.jpg",
      prompt="Add Chinese New Year decorations and red lanterns to the background",
      preserve_subject=True
  )

Prompt Best Practices

Basic Formula

[Subject] + [Action/Pose] + [Scene/Background] + [Style] + [Atmosphere/Lighting]

Style Keywords

• Realistic: photorealistic, hyperrealistic, 8K, detailed
• Anime: anime style, Ghibli style, cel shading
• Oil Painting: oil painting style, impressionist, Van Gogh style
• Minimalist: minimal, flat design, vector art
• Cinematic: cinematic, dramatic lighting, movie poster style

Quality Enhancement Words

• high quality

  ,

  detailed

  ,

  sharp focus

• professional photography

  ,

  award winning

• 4K resolution

  ,

  ultra detailed

Avoidance Items

• ❌ Avoid vague descriptions: "A nice picture"
• ❌ Avoid contradictory descriptions: "Cartoon in realistic style"
• ❌ Avoid sensitive content
• ✅ Be specific, clear, and layered

Application Scenario Templates

Scenario 1: WeChat Red Envelope Cover/Festival Greeting Image

prompt_template: |
  A {animal} in {pose}, surrounded by {decorations}, 
  Chinese New Year theme, festive red and gold colors, 
  {style} style, high quality, {text_content}
  
variables:
  animal: "majestic horse" # Year of the Horse
  pose: "running gracefully"
  decorations: "cherry blossoms, red lanterns, gold coins"
  style: "elegant illustration"
  text_content: "with Chinese text '恭喜发财' in golden calligraphy"

Scenario 2: Presentation Illustration

prompt_template: |
  {concept} visualization, professional infographic style,
  clean white background, modern corporate aesthetic,
  subtle gradients, minimalist design

variables:
  concept: "AI workflow automation"

Scenario 3: Social Media Content

prompt_template: |
  {subject} {action}, {platform} optimized aspect ratio,
  vibrant colors, eye-catching composition, 
  trending aesthetic, shareable content style
  
variables:
  subject: "coffee cup"
  action: "with steam rising"
  platform: "Instagram" # 1:1 or 4:5

Output Format

Generated Result Display

## 🎨 Image Generation Completed

**Prompt**: [English prompt used]

**Parameters**:
- Style: [style]
- Aspect Ratio: [aspect_ratio]
- Resolution: [resolution]

**Generated Results**:
![Generated Image](path/to/image.png)

**Next Steps**:
- [ ] Satisfied, save to specified location
- [ ] Need to adjust style/color
- [ ] Need to modify specific parts
- [ ] Regenerate

Notes

1. Copyright Compliance: Generated images have SynthID watermarks
2. Content Policy: Comply with Google's usage policies, do not generate sensitive content
3. Commercial Use: Supports commercial use (marketing, products)
4. Text Rendering: Nano Banana Pro supports multilingual text, but Chinese effects need verification
5. Character Consistency: Maintaining character features across images requires using the reference image function

Resource References

• resources/prompt_templates.yaml

  - Preset prompt templates

• resources/style_presets.md

  - Detailed style presets

• resources/chinese_new_year_2026.md

  - Year of the Horse exclusive templates