banana-ads

name	banana-ads
description	Generate advertising visuals — banners, social media ads, display ads, and promotional graphics with auto-prompt building and format-aware aspect ratios. Triggers on: ad, banner, advertisement, display ad, social ad, promo graphic, campaign visual, marketing image, sponsored post image, ad creative.
allowed-tools	Bash(uv:*), Bash(ls:*), Read

Online Advertising Visuals

Input: $ARGUMENTS (optional — product/service and ad format)

If $ARGUMENTS is provided, use it as the campaign brief and skip to prompt building. If empty, ask about the product and target placement.

Requirements

• uv installed
• GEMINI_API_KEY environment variable set (get one at https://aistudio.google.com/apikey)

Workflow

1. Understand — Clarify the product or service, the value proposition, and the target ad placement (Instagram feed, YouTube pre-roll, web banner, etc.).
2. Build prompt — Apply the Prompt Formula below to construct a bold, commercially viable image description.
3. Configure — Detect the ad format from context and select aspect ratio and resolution from the Ad Format Detection table.
4. Generate — Run the script. the generation script with the assembled prompt and settings.
5. Deliver — Report the saved file path. Do NOT read the image file back. Offer to iterate: adjust composition, color palette, or layout.

Prompt Formula

[Product/Service] + [Value Proposition as Action] + [Ad Placement Context] + [Bold Center-Weighted Composition] + [Clean Commercial Photography or Flat Design]

Ad Format Detection

Placement	Aspect Ratio	Resolution
Instagram/Facebook feed	4:5	2K
Instagram story	9:16	2K
YouTube pre-roll	16:9	2K
Web banner	21:9	2K
Square social	1:1	2K

Key Instructions

• Negative space for text overlay -- Always leave generous negative space (top, bottom, or side) for headline and CTA placement. Ad visuals are incomplete without copy; design the composition to accommodate it.
• No baked-in text -- Do NOT prompt Gemini to render text into the image. Gemini text rendering is unreliable and produces artifacts. Advise the user to add headlines, CTAs, and legal copy in a separate design tool (Figma, Canva, Photoshop).
• Bold center-weighted composition -- Place the hero product or key visual element at the center or slightly off-center with clean breathing room around it.
• Commercial lighting -- Default to clean, bright studio lighting with soft shadows. Use high-key lighting for upbeat campaigns and moody side lighting for premium or luxury positioning.
• Color psychology -- Match the color palette to the campaign intent: warm tones for urgency and sales, cool tones for trust and tech, vibrant saturated hues for youth and energy.

Generation

uv run "${CLAUDE_PLUGIN_ROOT}/scripts/banana.py" \
  --prompt "<detailed prompt following the formula>" \
  --output "<YYYY-MM-DD-HH-MM-SS-descriptive-name>.png" \
  --resolution <2K> \
  --aspect-ratio <ratio from Ad Format Detection table> \
  --model gemini-3-pro-image-preview

Editing (with input images)

When the user provides a product photo or existing visual to modify:

uv run "${CLAUDE_PLUGIN_ROOT}/scripts/banana.py" \
  --prompt "<edit instruction: what to change AND what to keep>" \
  --output "<YYYY-MM-DD-HH-MM-SS-descriptive-name>.png" \
  --input-image "/path/to/source.png" \
  --aspect-ratio <ratio>

For edits, describe both the change and what must stay identical:

"Place this product on a clean bright studio background with soft shadow. Keep the product, its color, and proportions exactly the same."

Up to 14 input images can be passed (repeat --input-image for each).

References

See references/prompts.md for ready-to-use example prompts covering common advertising visual scenarios.