Nano Banana — Simple Graphic Generator

• 1920x1080 — landscape (default)
• 1080x1920 — portrait / mobile
• 1080x1080 — square / social media
• 1200x628 — OG image / link preview
• Or any custom dimensions the user provides

If no size is mentioned, default to 1920x1080 landscape.

Step 2: Build the prompt

Write a detailed image generation prompt that tells Gemini exactly what to create. A good prompt includes:

• Dimensions: Use the size the user requested, or 1920x1080 landscape by default
• Visual style: Default to a clean modern style — white or light background, flat vector illustrations, geometric sans-serif text, bold accent colors. Override if the user requested something specific.
• Content: The key information laid out clearly — headings, bullet points, icons, or simple charts as appropriate
• Layout direction: How sections should flow (left-to-right columns, grid, top-to-bottom)

Save the prompt to a temporary file for the script to read.

Example prompt structure:

Generate a [WIDTHxHEIGHT] [orientation] infographic image.

Topic: [topic]
Style: Modern flat design with white background, corporate blue (#3B5998) headings,
accent colors (#FF6B6B, #FFD93D, #6C5CE7). Clean geometric sans-serif typography.

Layout: [describe the layout — e.g., "3 columns flowing left to right"]

Content:
- Title: [title]
- Section 1: [heading + key points]
- Section 2: [heading + key points]
- Section 3: [heading + key points]

Include flat vector icons next to each section heading.
Add subtle decorative elements (dots, wavy lines) in the background.
All text must be clearly readable.

Step 3: Generate the image

Run the generation script:

python3 <skill-path>/scripts/generate_image.py \
  --prompt-file <prompt-file> \
  --output <output-dir>/graphic.png \
  --model gemini-2.5-flash-image

For higher quality results (at the cost of speed), use --model gemini-3-pro-image-preview.

Step 4: Present the result

Show the generated image to the user. Offer:

• Regenerate with a different style or layout
• Refine by editing the prompt and running again
• Try a different model if quality isn't satisfactory

Tips for good results

• Keep text content short — Gemini handles brief bullet points better than dense paragraphs
• Limit to 3-5 sections for clarity
• Explicit color codes in the prompt help maintain consistency
• If the first result isn't great, try rephrasing the prompt rather than adding more constraints