nanobanana

This skill uses the nanobanana CLI (Gemini image editing) to apply prompt-based edits to images.

When to Use

• The user wants an image edited (remove objects, change background, stylize, improve, etc.).
• The workflow needs a real output file (PNG/JPEG) to show or further process.

Requirements

• nanobanana CLI installed via Nix (no global pip/uv installs).
• GEMINI_API_KEY provided via agenix (/run/agenix/gemini-api-key). The CLI wrapper auto-loads this if the env var is not set.

Workflow

1. Ask for the input image path and the edit prompt.
2. If no output path is provided, let the CLI generate <input>_edited.<ext>.
3. Run:

   nanobanana <image-path> "<prompt>" [output-path]
   

4. Confirm the output file exists and report its path.

Notes

• The model may upsample images (outputs are often 1024x1024 or larger).
• Keep prompts short and specific. If the result is off, iterate with a refined prompt.
• Use /tmp for temporary inputs and outputs unless the user requests otherwise.