Banana Claude: generate and edit Gemini images inside Claude Code

Banana Claude is a free, open-source plugin for Claude Code by Agrici Daniel. It connects Claude Code to Google's Gemini image models so you can create and edit images without leaving your terminal.

The interesting part is how it handles the gap between "I want an image" and "Gemini, here's my API call." Instead of firing off your one-line description and hoping it works, Banana Claude plans the image first. It writes a full creative brief, shows you the prompt it intends to send, tells you which model it picked and what it will cost, and waits. Nothing goes to Google until you say yes.

Why it's worth installing

I went through the project repository and three things stood out:

You approve every request before it costs anything. The planning step runs locally. No prompt is sent to Gemini, and nothing is billed, until you review the plan and confirm (README.md).

It installs clean. The whole thing runs on standard Python with no extra packages. No dependency tangles, no venv surprises (CLAUDE.md).

Your API key stays out of logs. The key is masked in command output and your cost history is stored locally on your machine (SECURITY.md).

How it works

The workflow is a six-step loop:

1. Ask. Describe the image you want in plain English.
2. Plan. Claude turns your short idea into a detailed creative brief and picks the Gemini model that fits.
3. Review. You see the final prompt, the chosen model, privacy settings, and an estimated cost.
4. Approve. One approval equals one image attempt. There are no hidden retries.
5. Create. The tool generates the image and saves it to your folder.
6. Check. Claude inspects the result for problems (bad cropping, garbled text, visual glitches) so you can fix it, regenerate, or ship it.

Which Gemini models does it use?

Banana Claude picks the model automatically based on the task (SKILL.md):

Prerequisites

You need four things before installing Banana Claude:

• Claude Code version 2.1.199 or newer
• Python 3.11 or newer
• A Gemini API key linked to a billing-enabled Google AI account
• Mac, Linux, or Windows with WSL (standard Windows without WSL is not supported yet)

Installation

Run these four commands inside Claude Code:

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

Then ask for an image:

/banana-claude:banana generate an urban 16:9 GitHub hero with clean left-side copy space

What you can do with it

With Banana Claude, you handle your visual workflow in one place:

• Generate from scratch. Blog covers, YouTube thumbnails, product shots, diagrams, social media graphics.
• Edit selectively. Change one part of an image while keeping the rest intact.
• Keep characters consistent. Continue a previous session or attach reference photos so a character or product looks the same across multiple shots.
• Compare models. Test up to 3 prompts across 3 Gemini models at once to compare results.
• Overlay text and logos locally. AI misspells words constantly. Banana Claude can add your exact text, fonts, and logos through local compositing, so the spelling is always right.
• Catch flaws before publishing. Claude reviews the final image for composition, cropping, and visual glitches.

How it fits into a Claude Code workflow

You stop switching windows. Your code, docs, and images all live in the same terminal session.

You also stop writing elaborate prompts. Give a one-liner and Claude fills in the camera angle, lighting direction, and style details for you.

Cost stays visible. You approve every request after seeing the price, so accidental generations don't happen. And because the tool can overlay local text and logos on top of AI-generated backgrounds, your visuals come out closer to production-ready without a separate design tool.