moai-connector-nano-banana

Original source / Raw SKILL.md

---
name: moai-connector-nano-banana
description: Professional image generation with Google Nano Banana Pro (Gemini 3 Pro)
version: 1.0.1
modularized: true
last_updated: 2025-11-30
compliance_score: 70
auto_trigger_keywords:
  - banana
  - connector
  - nano
color: red
---

## 🎯 Quick Reference (30 seconds)

**Purpose**: Professional image generation using Google's Nano Banana Pro (Gemini 3 Pro Image Preview).

**Key Features**:
- **Text-to-Image**: Detailed prompts → 1K/2K/4K resolution images
- **Image-to-Image**: Style transfer, object manipulation, editing
- **Real-time Grounding**: Google Search integration for factual content
- **Multi-Reference**: Up to 14 reference images (6 objects + 5 humans)
- **Advanced Text**: Sophisticated text rendering directly in images

**Two Models**:
1. **Nano Banana Pro** (gemini-3-pro-image-preview) - Professional quality, 10-60s
2. **Gemini 2.5 Flash** (gemini-2.5-flash-image) - Fast, ~5-15s

---


## Implementation Guide (5 minutes)

### Features

- Text-to-Image generation with 1K/2K/4K resolutions
- Image-to-Image editing and style transfer
- Multi-turn refinement for iterative improvements
- Reference image guidance (up to 14 references)
- Real-time Google Search grounding for factual content
- Advanced text rendering directly in images

### When to Use

- Generating professional visual assets for documentation or marketing
- Creating UI mockups and design concepts quickly
- Producing social media graphics and promotional images
- Illustrating technical documentation with custom diagrams
- Rapid prototyping of visual ideas before final design work

### Core Patterns

**Pattern 1: Structured Prompt for Quality**
```python
prompt = """
A serene Japanese garden at golden hour.
Lighting: warm sunset light filtering through maple trees.
Camera: wide-angle 35mm lens, low angle shot.
Composition: Rule of thirds, stone path leading to pagoda.
Color palette: warm gold, jade green, soft cream.
Style: photorealistic with slight cinematic color grading.
Quality: 4K resolution. Final output: PNG.
"""
```

**Pattern 2: Multi-Turn Refinement**
1. Generate initial image with base prompt
2. Review output and identify areas for improvement
3. Provide targeted refinement: "Make sky more dramatic"
4. Iterate up to 5 turns for perfect result

**Pattern 3: Reference-Guided Generation**
```python
# Use reference images to guide style
generate_image(
    prompt="Mountain landscape in the style of reference",
    reference_images=["style_ref.png", "composition_ref.png"],
    resolution="2K",
    aspect_ratio="16:9"
)
```

## 📚 Core Patterns (5-10 minutes)

### Pattern 1: Prompt Structure for Quality Images

**Key Concept**: Well-structured prompts generate better images

**Template**:
```
[Scene Description]
A [adjective] [subject] doing [action].
Setting: [location] with [environmental details].

[Photographic Elements]
Lighting: [type], creating [mood].
Camera: [angle] shot with [lens] lens.
Composition: [framing_details].

[Color & Style]
Color palette: [colors]. Style: [art_style].
Quality: [resolution]. Final output: [format].
```

**Example**:
```
A serene Japanese garden at golden hour.
Lighting: warm sunset light, creating peaceful mood.
Camera: wide-angle 35mm lens shot.
Color palette: gold, jade green, cream.
Quality: 4K photorealistic. Final: PNG.
```

### Pattern 2: Text-to-Image Generation

**Key Concept**: Generate professional images from text prompts

**Basic Flow**:
1. Write detailed, structured prompt
2. Choose resolution (1K, 2K, 4K)
3. Select aspect ratio (1:1, 16:9, 3:2, etc.)
4. Enable Google Search for current information (optional)
5. Generate and retrieve Base64 PNG

**Execution**:
```python
image_data = generate_image(
    prompt="Your detailed prompt here",
    resolution="2K",
    aspect_ratio="16:9",
    enable_google_search=True,  # For current info
    thinking_process=True        # Auto-optimize
)
```

### Pattern 3: Image-to-Image Editing

**Key Concept**: Transform existing images with detailed instructions

**Common Tasks**:
- **Style Transfer**: Convert to art style (Van Gogh, anime, etc.)
- **Object Manipulation**: Add, remove, or modify elements
- **Composition Change**: Reframe, zoom, or reposition subjects
- **Quality Enhancement**: Upscale, improve detail, adjust colors

**Flow**:
1. Load original image
2. Write transformation instruction
3. Reference images (optional)
4. Apply edit maintaining coherence
5. Retrieve edited image

### Pattern 4: Multi-Turn Refinement

**Key Concept**: Iteratively improve images through conversation

**Workflow**:
1. Generate initial image
2. Review output
3. Provide refinement instruction
4. Regenerate with improvements
5. Repeat (max 5 turns)

**Example**:
```
Turn 1: "A mountain landscape at sunset"
Turn 2: "Make the sky more dramatic with purple clouds"
Turn 3: "Add a lone tree in foreground"
```

### Pattern 5: Reference Image Guidance

**Key Concept**: Use reference images to guide generation style

**Supported References**:
- Up to 6 object references
- Up to 5 human references
- Style influences
- Composition guides

**Usage**:
```python
generate_image(
    prompt="Similar style to reference",
    reference_images=[
        "path/to/style_reference.png",
        "path/to/composition_ref.png"
    ]
)
```

---

## 📖 Advanced Documentation

This Skill uses Progressive Disclosure. For detailed implementation:

- **[modules/prompt-engineering.md](modules/prompt-engineering.md)** - Professional prompt templates
- **[modules/api-reference.md](modules/api-reference.md)** - Complete API documentation
- **[modules/examples.md](modules/examples.md)** - Real-world usage examples
- **[modules/troubleshooting.md](modules/troubleshooting.md)** - Common issues and solutions

---

## 🎨 Model Selection Guide

**Choose Nano Banana Pro when**:
- Professional quality required
- 2K/4K resolution needed
- Complex compositions
- Sophisticated text in images
- Real-time information important
- Budget allows (higher cost)

**Choose Gemini 2.5 Flash when**:
- Quick iterations needed
- Prototyping and testing
- High volume generation
- 1K resolution sufficient
- Speed critical
- Cost sensitive

---

## Quick Reference (30 seconds)

**Core Purpose**: Professional AI image generation using Nano Banana Pro (Gemini 3 Pro) and Gemini 2.5 Flash.

**Key Features**: Text-to-image, image-to-image editing, multi-turn refinement, reference guidance, 4K resolution.

**When to Use**: Visual asset creation, prototyping, documentation, UI mockups, marketing materials.

---

## Works Well With

**Agents**:
- **design-uiux** - UI/UX design integration
- **code-frontend** - Frontend asset implementation
- **workflow-docs** - Visual documentation generation

**Skills**:
- **moai-lang-unified** - UI/UX implementation with generated assets
- **moai-docs-generation** - Create visual documentation
- **moai-cc-claude-md** - Embed generated images in markdown
- **moai-domain-frontend** - Frontend integration

**Commands**:
- `/moai:3-sync` - Documentation with visual assets
- `/moai:9-feedback` - Image generation improvements

---

## 🔗 Integration with Other Skills

**Typical Workflow**:
1. Use this Skill to generate visual assets
2. Use moai-domain-frontend to implement in UI
3. Use moai-docs-generation to document with images

---

## 📈 Version History

**1.0.1** (2025-11-23)
- 🔄 Refactored with Progressive Disclosure pattern
- 📚 Detailed prompts moved to modules/
- ✨ Core patterns highlighted in SKILL.md
- ✨ Added model selection guide

**1.0.0** (2025-11-12)
- ✨ Nano Banana Pro (Gemini 3 Pro) support
- ✨ Text-to-Image and Image-to-Image
- ✨ Multi-turn refinement capability
- ✨ Reference image guidance

---

**Maintained by**: alfred
**Domain**: Image Generation & Visual Creation
**Generated with**: MoAI-ADK Skill Factory