Nano Banana CLI - Image Generation Skill

Optional flags:

• --out <path>: Save to a specific file path (for single image output)
• --out-dir <dir>: Directory for multiple output images (default: current dir)
• --aspect-ratio <ratio>: e.g., 1:1, 16:9, 9:16
• --image-size <size>: e.g., 1K
• --modalities <types>: Output types, comma-separated: image or image,text (default: image,text)
• --stream: Use streaming mode (SSE delta aggregation)
• --raw-json: Print raw OpenRouter API response
• --timeout <duration>: Request timeout (default: 2m)
• --referer <url>: HTTP Referer header
• --title <string>: X-Title header

2. Edit an Existing Image

Pass one or more input images along with a prompt describing the edit:

nano-banana-cli generate \
  --model google/gemini-2.5-flash-image-preview \
  --input-image ./source.png \
  --prompt "Make the background a sunset" \
  --out edited.png

You can pass multiple --input-image flags for multi-image editing. Input images can be:

• Local file paths (e.g., ./photo.jpg)
• HTTP/HTTPS URLs
• Data URLs

3. List Available Models

nano-banana-cli models
nano-banana-cli models --format json

Returns models that support image output. Use --format json for structured output.

Workflow Guide

When the user asks you to generate an image, follow this process:

1. Check the environment: Verify OPENROUTER_API_KEY is set. If not, ask the user.
2. Build the CLI if not already built: go build -o nano-banana-cli .
3. Pick a model: If the user didn't specify a model, use google/gemini-2.5-flash-image-preview as a good default. You can also run nano-banana-cli models to see all available options.
4. Craft the prompt: Use the user's description directly or enhance it for better results.
5. Choose output location: Use --out for a specific filename, or --out-dir for batch output.
6. Run the command and report the output file path to the user.

Common Patterns

Generate with specific aspect ratio

nano-banana-cli generate \
  --model google/gemini-2.5-flash-image-preview \
  --prompt "A panoramic mountain landscape" \
  --aspect-ratio 16:9 \
  --out landscape.png

Image-only output (no text)

Some models (like flux) only support image output:

nano-banana-cli generate \
  --model black-forest-labs/flux.2-pro \
  --modalities image \
  --prompt "Abstract art" \
  --out art.png

Streaming mode for long generations

nano-banana-cli generate \
  --model google/gemini-2.5-flash-image-preview \
  --stream \
  --prompt "A detailed fantasy castle" \
  --out castle.png

Error Handling

• If OPENROUTER_API_KEY is missing, the CLI will exit with an error. Prompt the user to set it.
• If --model is missing, the CLI will error. Always include a model flag.
• Network timeouts default to 2 minutes. For complex generations, consider increasing with --timeout 5m.
• If --out and --out-dir are both specified, the CLI will error (they're mutually exclusive).

Tips

• The default modalities are image,text which works with most models. Some models (like Flux) only support image.
• Generated images are saved as PNG by default. The file extension in --out doesn't affect the actual format — it depends on the model's output.
• When editing images, the prompt should describe what to change, not the full image.