nano-banana-2

Nano Banana 2 Image Generation Skill

Generate images from text prompts using Google's Nano Banana 2 model with web search grounding for accurate, contextually-aware visual content.

Overview

This skill integrates with Google's Gemini API to generate images using the Nano Banana 2 model (gemini-3.1-flash-image-preview), optimised for speed and high-volume use.

Key Capabilities:

• Text-to-image generation from natural language prompts
• Web search grounding for real-world accuracy (enabled by default)
• Configurable aspect ratios (1:1, 16:9, 4:5, 9:16, etc.)
• Multiple image sizes (512px, 1K, 2K, 4K)
• Integration with other SAS-AM skills (presentations, social media, research)

Use Cases:

• Presentation visuals for sas-presentation
• Social media graphics for linkedin-post-generator
• Industry/market infographics for b2b-research-agent
• Proposal visuals for beam-selling
• Standalone creative image generation

---

⚠️ MANDATORY: Post-Processing Required

Every generated image MUST be post-processed before delivery.

After saving the raw PNG, ALWAYS run:

node ~/.claude/SASAMClaudeCodeSkills/nano-banana-2/1.1.0/scripts/post-process.js <input.png> <output_final.jpg>

This adds the SAS logo watermark and optimises the image. The _final.jpg file is what you deliver to the user, NOT the raw PNG.

DO NOT:

• Add "watermark" or "SAS logo" to the generation prompt
• Deliver the raw PNG without post-processing
• Skip this step for any reason

---

Commands

Command	Action
setup	Run first-time API key configuration
status	Show current configuration and statistics
generate [prompt]	Generate an image from text prompt
history	Show recent generation history
[prompt]	Shorthand for generate (most common usage)

Invocation Examples

/nano-banana-2 setup                                    # First-time configuration
/nano-banana-2 status                                   # Show config and stats
/nano-banana-2 "A futuristic city skyline at sunset"   # Generate image
/nano-banana-2 "Professional infographic about AI trends" --aspect 16:9 --size 2K
/nano-banana-2 history                                  # Recent generations

---

Discovery Process

Check Configuration State

Before generating any image, check if ~/.claude/skills/nano-banana-2/config.json exists.

If config exists:

1. Load configuration from ~/.claude/skills/nano-banana-2/config.json
2. Validate API key is set
3. Proceed with generation using saved preferences

If config does NOT exist:

1. Run first-time setup interview
2. Guide through API key configuration
3. Set default preferences
4. Create config file
5. Then proceed with generation

First-Time Setup Questions

When no configuration exists, ask:

1. API Key

   • Do you have a Google AI Studio API key?
   • If yes: Ask for the key
   • If no: Guide to https://aistudio.google.com/apikey
   • Validate key with a test API call before saving

2. Default Aspect Ratio

   • What's your typical use case?
   • Options: 16:9 (presentations/videos), 1:1 (social media), 4:5 (portrait/mobile)

3. Web Search Grounding

   • Enable web search grounding by default? (Recommended: Yes)
   • This helps generate more accurate, contextually-aware images

4. Output Directory

   • Use default ./generated-images/ or specify custom location?

Setup Guide for API Key

If the user doesn't have an API key, provide these instructions:

To get your Google AI Studio API key:

1. Go to https://aistudio.google.com/apikey
2. Sign in with your Google account
3. Click "Create API Key"
4. Copy the generated key (starts with "AIza...")
5. Paste it here when prompted

Note: The API key will be stored globally in ~/.claude/skills/nano-banana-2/config.json
This location is outside your project directory, so credentials are protected.

---

Configuration State

Store configuration in ~/.claude/skills/nano-banana-2/config.json:

{
  "schema_version": "1.0.0",
  "created_at": "2026-02-27T10:30:00Z",
  "updated_at": "2026-02-27T10:30:00Z",
  "api_key": "AIza...",
  "verified_at": "2026-02-27T10:30:00Z",
  "preferences": {
    "default_aspect_ratio": "16:9",
    "default_image_size": "2K",
    "web_search_enabled": true,
    "output_directory": "./generated-images"
  },
  "statistics": {
    "total_generated": 0,
    "last_generated_at": null
  }
}

See references/config-template.json for the full schema.

Security:

• Config is stored in ~/.claude/skills/nano-banana-2/ (outside project directory)
• No .gitignore changes needed - credentials are not in the repo
• When displaying status, mask the API key: AIza...xyz

---

Generating Images

API Call Structure

Use the Bash tool to call the Gemini API via curl:

API_KEY="{{API_KEY}}"
PROMPT="{{USER_PROMPT}}"
ASPECT_RATIO="{{ASPECT_RATIO}}"  # e.g., "16:9"
IMAGE_SIZE="{{IMAGE_SIZE}}"      # e.g., "2K"

curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-flash-image-preview:generateContent" \
  -H "Content-Type: application/json" \
  -H "x-goog-api-key: ${API_KEY}" \
  -d '{
    "contents": [{"parts": [{"text": "'"${PROMPT}"'"}]}],
    "generationConfig": {
      "responseModalities": ["IMAGE"],
      "imageConfig": {
        "aspectRatio": "'"${ASPECT_RATIO}"'",
        "imageSize": "'"${IMAGE_SIZE}"'"
      }
    },
    "tools": [{"googleSearch": {}}]
  }'

Response Handling

The API returns JSON with base64-encoded image data:

{
  "candidates": [{
    "content": {
      "parts": [{
        "inlineData": {
          "mimeType": "image/png",
          "data": "iVBORw0KGgo..."
        }
      }]
    }
  }]
}

Processing Steps:

1. Parse JSON response
2. Extract base64 data from candidates[0].content.parts[0].inlineData.data
3. Decode base64 to binary
4. Save as PNG file with timestamp and prompt summary
5. Create metadata JSON file alongside image

Extracting and Saving Image

# Parse response and save image
RESPONSE=$(curl ... )  # API call from above

# Extract base64 data (using jq)
IMAGE_DATA=$(echo "$RESPONSE" | jq -r '.candidates[0].content.parts[0].inlineData.data')

# Generate filename with timestamp and prompt summary
TIMESTAMP=$(date +%Y-%m-%d_%H%M%S)
PROMPT_SLUG=$(echo "$PROMPT" | tr '[:upper:]' '[:lower:]' | sed 's/[^a-z0-9]/-/g' | cut -c1-40)
FILENAME="${TIMESTAMP}_${PROMPT_SLUG}"

# Create output directory if needed
mkdir -p ./generated-images

# Save raw image
echo "$IMAGE_DATA" | base64 -d > "./generated-images/${FILENAME}.png"

# MANDATORY: Run post-processing to add watermark
node ~/.claude/SASAMClaudeCodeSkills/nano-banana-2/1.1.0/scripts/post-process.js "./generated-images/${FILENAME}.png" "./generated-images/${FILENAME}_final.jpg"

# Save metadata (reference the final watermarked file)
cat > "./generated-images/${FILENAME}.json" << EOF
{
  "prompt": "${PROMPT}",
  "aspect_ratio": "${ASPECT_RATIO}",
  "image_size": "${IMAGE_SIZE}",
  "web_search_enabled": true,
  "generated_at": "$(date -u +%Y-%m-%dT%H:%M:%SZ)",
  "raw_file": "./generated-images/${FILENAME}.png",
  "final_file": "./generated-images/${FILENAME}_final.jpg"
}
EOF

IMPORTANT: The _final.jpg file is the deliverable. Always provide this file path to the user, not the raw PNG.

---

Web Search Grounding

Web search grounding enhances image generation by allowing the model to reference real-world information.

How It Works:

• When enabled, the model can search the web for context about the prompt
• Results in more accurate depictions of real entities, current events, and factual details
• Particularly useful for generating images of real products, places, or people

When to Use:

• Generating images of real-world subjects (cities, landmarks, products)
• Creating infographics with factual content
• Producing visuals that reference current trends or events

When to Disable:

• Pure creative/fictional imagery
• Abstract art
• When you want the model to rely solely on its training data

Disabling for a Single Request:

If the user specifies --no-search or asks for purely creative imagery, modify the API call to omit the tools parameter:

# Without web search grounding
curl -s -X POST \
  "https://generativelanguage.googleapis.com/v1beta/models/gemini-3.1-flash-image-preview:generateContent" \
  -H "Content-Type: application/json" \
  -H "x-goog-api-key: ${API_KEY}" \
  -d '{
    "contents": [{"parts": [{"text": "'"${PROMPT}"'"}]}],
    "generationConfig": {
      "responseModalities": ["IMAGE"],
      "imageConfig": {
        "aspectRatio": "'"${ASPECT_RATIO}"'",
        "imageSize": "'"${IMAGE_SIZE}"'"
      }
    }
  }'

---

Parameter Reference

Aspect Ratios

Ratio	Best For
1:1	Social media posts, profile images, thumbnails
16:9	Presentations, video thumbnails, desktop wallpapers
9:16	Mobile wallpapers, stories, vertical video
4:5	Portrait images, Instagram posts
4:3	Traditional photo format
3:4	Portrait traditional format

Image Sizes

Size	Dimensions (approx)	Best For
512px	512x512 (or aspect equivalent)	Quick previews, thumbnails
1K	~1024px	Web use, presentations
2K	~2048px	High-quality prints, detailed work (default)
4K	~4096px	Large prints, maximum detail

---

Output Management

File Naming Convention

./generated-images/
├── 2026-02-27_143022_futuristic-city-skyline-at-sunset.png
├── 2026-02-27_143022_futuristic-city-skyline-at-sunset.json
├── 2026-02-27_145530_professional-infographic-about-ai-t.png
├── 2026-02-27_145530_professional-infographic-about-ai-t.json
└── ...

Metadata File Structure

Each generated image has an accompanying .json file:

{
  "prompt": "A futuristic city skyline at sunset",
  "aspect_ratio": "16:9",
  "image_size": "2K",
  "web_search_enabled": true,
  "generated_at": "2026-02-27T14:30:22Z",
  "file_path": "./generated-images/2026-02-27_143022_futuristic-city-skyline-at-sunset.png",
  "model": "gemini-3.1-flash-image-preview"
}

After Generation

CRITICAL: Always run post-processing after saving the raw PNG.

1. Run post-processing script (REQUIRED):

   node ~/.claude/SASAMClaudeCodeSkills/nano-banana-2/1.1.0/scripts/post-process.js <input.png> <output.jpg>
   

2. Report success with the final watermarked file path
3. Update statistics in config file
4. Offer next steps (regenerate, adjust parameters, use in another skill)

---

Post-Processing

All generated images MUST be post-processed to add the SAS watermark and optimise for web use.

Post-Processing Script

Run this after every image generation:

node ~/.claude/SASAMClaudeCodeSkills/nano-banana-2/1.1.0/scripts/post-process.js ./generated-images/IMAGE.png ./generated-images/IMAGE_final.jpg

The script automatically:

• Resizes to 1000px (maintaining aspect ratio)
• Analyzes corner brightness to select appropriate watermark
• Adds SAS logo watermark (bottom-right)
• Compresses to JPG with MozJPEG (quality 85)
• Outputs file under 200KB

Target Specifications

Property	Target
Dimensions	1000x1000 pixels (or aspect-equivalent)
File Size	Under 200KB
Format	JPG (MozJPEG optimised)
Watermark	SAS logo, bottom-right corner

Watermark Selection

The post-processing script automatically selects the appropriate watermark:

• Dark backgrounds (brightness < 128): Light SAS logo
• Light backgrounds (brightness >= 128): Dark SAS logo

Assets located at:

~/.claude/SASAMClaudeCodeSkills/nano-banana-2/1.1.0/skills/nano-banana-2/assets/
├── sas-logo-dark.png   # For light backgrounds
└── sas-logo-light.png  # For dark backgrounds

---

Integration with Other Skills

This skill can be called from other SAS-AM skills to generate visual content.

Integration Pattern

When another skill needs an image:

--- Generating Visual ---

[Skill Name] is generating an image using Nano Banana 2...

Prompt: [derived prompt]
Aspect Ratio: [contextually appropriate ratio]

Skill-Specific Integrations

Source Skill	Trigger Context	Suggested Parameters
sas-presentation	"Generate visual for slide about [topic]"	16:9, 2K
linkedin-post-generator	"Create image for LinkedIn post"	1:1 or 4:5, 2K
b2b-research-agent	"Generate infographic for [company/market]"	16:9 or 1:1, 2K
beam-selling	"Create visual for proposal section"	16:9, 2K

Cross-Skill Example

When sas-presentation needs a visual:

1. sas-presentation identifies need for visual asset
2. Check if nano-banana-2 is configured
3. Generate appropriate prompt based on slide content
4. Call nano-banana-2 with 16:9 aspect ratio
5. Save image to presentation assets directory
6. Reference in slide HTML

---

Error Handling

API Key Not Configured

No API key configured

Run `/nano-banana-2 setup` to configure your Google AI Studio API key.

To get an API key:
1. Go to https://aistudio.google.com/apikey
2. Sign in with your Google account
3. Create and copy your API key

Invalid API Key

API key validation failed

Error: Invalid API key or key has been revoked

To fix:
1. Verify your key at https://aistudio.google.com/apikey
2. Run `/nano-banana-2 setup` to enter a new key

Quota Exceeded

API quota exceeded

Your Google AI Studio quota has been reached.

Options:
- Wait for quota reset (typically daily)
- Upgrade your Google AI Studio plan
- Check quota usage at https://aistudio.google.com

Current statistics:
- Images generated today: [N]
- Last generation: [timestamp]

Content Policy Violation

Content policy restriction

The requested image could not be generated due to content policy.

Suggestions:
- Rephrase the prompt to be less specific about restricted content
- Remove references to real people, trademarks, or sensitive topics
- Focus on abstract or general descriptions

Original prompt: [prompt]

Network Error

Network error

Unable to reach the Google API endpoint.

Troubleshooting:
- Check your internet connection
- Verify firewall/proxy settings allow access to googleapis.com
- Try again in a few moments

If the issue persists, check Google Cloud status: https://status.cloud.google.com

Malformed Response

Unexpected API response

The API returned an unexpected response format.

This may indicate:
- A temporary API issue
- An API version change

Action: Retry the request. If the issue persists, report it.

Response received: [truncated response]

---

Status Command Output

When the user runs /nano-banana-2 status:

Nano Banana 2 Configuration
===========================

API Key: AIza...xyz (verified 2026-02-27)

Preferences:
  Aspect Ratio: 16:9
  Image Size: 2K
  Web Search: Enabled
  Output Dir: ./generated-images

Statistics:
  Total Generated: 42
  Last Generated: 2026-02-27 14:30:22

Recent Generations:
  1. futuristic-city-skyline-at-sunset.png (16:9, 2K)
  2. professional-infographic-about-ai-t.png (16:9, 2K)
  3. abstract-data-visualisation.png (1:1, 2K)

---

Pre-Delivery Checklist

Before completing a generation request, verify:

• Configuration loaded from ~/.claude/skills/nano-banana-2/config.json
• API key is valid and not expired
• Prompt is not empty
• Output directory exists or was created
• Image was successfully saved
• Metadata JSON was created alongside image
• Statistics updated in config

After generation:

• Report file path to user
• Display generation parameters used
• Offer to open image or generate variations
• Update last_generated_at timestamp

---

Content Guidelines

Prompt Engineering Tips:

• Be specific and descriptive
• Include style keywords (photorealistic, digital art, watercolour, minimalist)
• Specify lighting, mood, and composition
• Reference time of day, weather, or environment
• For infographics, describe the data story and visual style

Hyper-Realistic Prompt Enhancement (Default)

For maximum photorealism, automatically enhance user prompts with these modifiers:

Core Hyper-Realistic Keywords:

• hyperrealistic photograph
• photorealistic, 8K resolution
• professional DSLR photography, Canon EOS R5
• RAW photo style
• shallow depth of field
• cinematic lighting

Prompt Enhancement Pattern:

When the user provides a prompt, prepend hyper-realistic modifiers:

User prompt: "Industrial factory with machinery failure"

Enhanced prompt: "Hyperrealistic photograph of an industrial factory
with machinery failure, photorealistic, 8K resolution, professional
DSLR photography Canon EOS R5, dramatic cinematic lighting, RAW photo
style, shallow depth of field"

When NOT to Enhance:

• User explicitly requests illustration, cartoon, or artistic style
• Abstract or conceptual imagery
• Infographics and data visualisations
• User includes --no-enhance flag

Examples of Effective Prompts:

"A photorealistic sunset over Sydney Harbour Bridge, warm golden light,
dramatic clouds, professional photography style"

"Minimalist infographic showing AI adoption trends, clean lines,
SAS-AM brand colours (blue and green accents), modern corporate style"

"Abstract data visualisation representing network connections,
glowing nodes, dark background, futuristic technology aesthetic"

SAS-AM Brand Integration:

• When generating for SAS-AM content, incorporate brand colours where appropriate
• SAS Blue: #002244
• SAS Green: #69BE28
• Keep visuals professional and suitable for asset management audience

---

Reference Files

• references/config-template.json - Configuration schema
• references/api-examples.md - Sample API calls and responses
• references/prompt-guide.md - Prompt engineering best practices