AgriciDaniel/banana-claude

Banana Claude – Claude‑side plugin for Google Gemini image generation

What it is – A Claude Code plugin that lets you write a natural‑language request and have the system plan, review, and (after an explicit approval) call Google’s Gemini image models (Nano Banana 2 Lite / 2 / Pro) to create the picture. The plugin shows you the full prompt, model choice, data settings and cost estimate before any paid request is sent.

How it works

1. Ask – You type a plain description, e.g. /banana-claude:banana generate an urban 16:9 GitHub hero with clean left‑side copy space.
2. Plan – The plugin builds a “visual brief”, selects a Gemini model, and prepares a cost estimate.
3. Review – You see the exact prompt, reference images, privacy flags and the estimated charge.
4. Approve – A one‑time approval token authorises a single request to the Gemini API.
5. Create – The image is generated; a metadata side‑car (privacy‑conscious) is saved alongside.
6. Check – You inspect the result, optionally edit, regenerate or ship it.

Key capabilities

• Generate any visual asset (campaign graphics, product scenes, diagrams, social media images, etc.).
• Edit a generated image with a single change while preserving identity, geometry and brand details.
• Continue work using stored “Flash” or “Pro” sessions, or reuse the last image as a fresh reference.
• Compare up to three prompts across three model routes in a single bounded portfolio.
• Review copy, crop, composition, artifacts, rights and provenance before approval.
• Typeset locally‑added copy, fonts and logos when exact text placement matters.

Safety & control built‑in

• The plugin is installed in a disabled state because image generation costs money.
• API keys never appear in commands, URLs or CI logs; they stay in Claude’s private settings.
• Every paid call shows a full plan and requires an explicit user approval.
• One approval = one provider attempt; changing the plan forces a new approval.
• Cost records are private and, by default, omit raw prompts.
• Generated images are never automatically marked production‑ready; the user must review them.

Installation (Claude Code)

/plugin marketplace add AgriciDaniel/banana-claude
/plugin install banana-claude@banana-claude-marketplace
/plugin enable banana-claude@banana-claude-marketplace
/reload-plugins

Prerequisites: Claude Code 2.1.199+, Python 3.11+, a billing‑enabled Google AI project with a Gemini API key.

Typical use‑cases

• Marketing teams needing quick hero images for landing pages or social posts.
• Designers who want an AI‑generated first draft they can edit and iterate on.
• Content creators comparing different visual concepts side‑by‑side before committing budget.
• Small businesses that want a controlled, cost‑visible way to produce custom graphics.

Limitations / caveats

• The underlying Gemini models are probabilistic; results may vary in spelling, geometry, or policy compliance.
• No guarantee of rights clearance or final invoice accuracy – those remain the user’s responsibility.
• Requires a paid Google Gemini API key; the plugin itself does not provide free credits.
• The plugin only works inside Claude Code; it is not a standalone CLI or web app.

Documentation & support

• User guide, prompt‑crafting tips, MCP tool reference, security notes, changelog and upgrade guide are all in the repo (docs/, skills/banana/).
• Contributions are welcomed via the usual pull‑request process; issues should be filed without exposing keys or private media.

License – MIT (see LICENSE).

---

Built by Agrici Daniel, AI Workflow Architect – see his blog, AI Marketing Hub and YouTube channel for more context.