Banana

This is a creative director layer over Google's Gemini image generation that refuses to pass raw user prompts through. Instead, it interprets your request, picks a domain mode (cinema, product, portrait, etc.), then constructs a detailed prompt using a five-component formula with real camera specs, brand names, and micro-details. It's opinionated about workflow: you must generate through /banana commands, it checks for brand presets in your home directory, and it handles safety filters with automatic rephrasing. The MCP setup assumes you have Gemini API access and want Claude to act as the prompt engineer rather than doing it yourself. Good if you want guardrails and consistent output structure, limiting if you prefer direct control over generation parameters.

skills.sh installs: 948
GitHub stars: 1.2K
last updated: Aug 21, 2026

View on GitHub

Install

$ npx -y skills add agricidaniel/banana-claude --skill banana --agent claude-code

Installs into .claude/skills of the current project.

Files

Banana Claude

Turn user intent into a frozen visual brief, compile the exact prompt, plan the request, obtain approval, execute through the bundled Gemini client, and inspect the actual pixels. The prompt is a control artifact, not the finished work.

Plugin command: /banana-claude:banana. The standalone install uses /banana and the direct scripts, without plugin MCP or plugin-managed secrets.