name: ai-image-tools description: Generate and edit images using either OpenAI GPT Image 1.5 or Google's Nano Banana Pro (Gemini 3 Pro Image). Use when the user asks to generate/create/edit/modify images. Supports image-to-image editing for both providers and optional mask-based inpainting for OpenAI.

AI Image Tools (OpenAI + Gemini)

One unified skill for image generation + editing, supporting:

• OpenAI: GPT Image 1.5 (generation + edits, optional mask inpainting)
• Gemini: Nano Banana Pro (Gemini 3 Pro Image) (generation + image-to-image edits)

Usage

Run from your current working directory so outputs save where you're working.

Generate (text → image)

uv run scripts/generate_image.py --prompt "A moody cinematic portrait of a golden retriever" --filename "out.png"

Pick a provider explicitly:

# OpenAI (GPT Image 1.5)
uv run scripts/generate_image.py --provider openai --prompt "..." --filename "out.png"

# Gemini (Nano Banana Pro)
uv run scripts/generate_image.py --provider gemini --prompt "..." --filename "out.png"

Edit (image → image)

uv run scripts/generate_image.py --prompt "Make it look like a watercolor painting" --filename "out.png" --input-image "input.png"

Mask-based inpainting (OpenAI only):

uv run scripts/generate_image.py --provider openai --prompt "A red balloon" --filename "out.png" --input-image "input.png" --mask "mask.png"

Provider Selection

• Default --provider auto:
  • uses OpenAI if OPENAI_API_KEY (or --openai-api-key) is available
  • otherwise uses Gemini if GEMINI_API_KEY (or --gemini-api-key) is available
• Set --provider openai or --provider gemini to force one.

API Keys

• OpenAI:
  • env: OPENAI_API_KEY
  • flag: --openai-api-key
• Gemini:
  • env: GEMINI_API_KEY
  • flag: --gemini-api-key

Options (Provider-Specific)

OpenAI options

• --quality low|medium|high (generation only; default medium)
• --size 1024x1024|1024x1536|1536x1024|auto (default 1024x1024)
• --background transparent|opaque|auto (generation only; default auto)
• --mask path/to/mask.png (edits only)

Gemini options

• --resolution 1K|2K|4K (default 1K)

Notes

• Output is always saved as PNG at --filename.
• Don’t read the output image back into the model unless explicitly requested.