cover-image-generation

Image Generation Skill

This skill generates an appropriate cover image for the radio show based on a prompt, using the Gemini 3 Flash Image model, with a fallback to the Gemini 3 Pro Image model ("Nano Banana Pro") if needed.

Requirements

• Python 3.10+
• google-genai Python package (>= 2.0.1)

Instructions

1. Generate an image (Recommended: use the metadata file directly):

   python3 skills/cover-image-generation/scripts/generate_image.py \
     --workspace ./workspace \
     --metadata ./workspace/data/show_notes.json
   

   Note: Using --metadata will automatically extract the show title and apply a random, high-quality prompt template. This is the preferred method.

   Alternative (Manual prompt):

   python3 skills/cover-image-generation/scripts/generate_image.py \
     --workspace ./workspace \
     --prompt "A prompt describing the image"
   

2. Output:

   • The image will be saved to {workspace}/images/cover.png.

Model

• Primary Model: gemini-3-flash-image-preview
• Fallback Model: gemini-3-pro-image-preview
• Resolution: 1:1 (default)

Prompting rules

When using the --metadata option, this skill uses a set of predefined prompt templates and selects one at random to generate cover images. It dynamically inserts the show title into the selected template.

• Example Prompt: "A professional podcast cover image for a show titled 'AI Talk Radio' on the 'AI Talk Radio' station. The design features the text 'AI Talk Radio' in a bold, stylish white font centered on the cover. The background is a vibrant purple with a textured water ripple effect that covers the entire frame, creating a dynamic and clean aesthetic."

If you choose to use the --prompt option instead, you must construct the prompt yourself. In that case, follow these rules:

Forbidden themes

• Do not ask for futuristic, cyberpunk or neon themes
• Do not include any text other than the show title