nano-banana_skill

Overview

This skill provides a compact Python scripting workflow for generating and editing images with Gemini models using uv. It emphasizes single-purpose heredoc scripts that run inline, declare dependencies, and save results to tmp/ for quick iteration. The default recommendation is to use the higher-quality Gemini Pro image model unless the user requests otherwise.

How this skill works

Write small Python snippets inside a heredoc and execute them with uv run, declaring dependencies in an inline metadata block. The scripts call the Google GenAI client to request image generation, configure image parameters (aspect ratio, size), and convert inline_data to PIL Images for saving. Use the same pattern to edit existing images by including the image object in the request.

When to use it

• Generate a new image, illustration, or AI art from a text prompt
• Edit or augment an existing image (add objects, change style, retouch)
• Run one-off Python image tasks without creating permanent files
• Prototype image generation workflows before making a reusable script
• Request both image and explanatory text in a single call

Best practices

• Keep each script focused on one task: generate, refine, or save
• Declare dependencies (google-genai, pillow) in the heredoc metadata
• Save outputs to tmp/ and print status for easy visual verification
• Default to gemini-3-pro-image-preview for quality, switch for cost or speed
• Iterate: run, inspect saved image, refine prompt, repeat

Example use cases

• Quickly generate a character illustration using an inline prompt and save to tmp/generated.png
• Add a hat or other element to an existing photo by loading the image and including it in the request
• Produce both an image and a short caption by requesting TEXT and IMAGE modalities
• Prototype a multi-step pipeline: generate draft, refine style, then export final asset
• Run a small image-processing utility (resize, format convert) via a one-off uv heredoc script