AgriciDaniel/banana-claude

Banana Claude – AI‑guided image creation for Gemini

What it does
Banana Claude is a command‑line / plugin tool that turns a natural‑language description of an image into a fully planned Gemini (Google AI) image‑generation request. It first builds a visual brief (prompt, model choice, data settings, cost estimate) that you can review before any paid call is made. After you approve the plan, it runs the request, saves the image together with a privacy‑focused metadata side‑car, and lets you inspect, edit, iterate or compare up to three variants.

Key features

• Plan‑first workflow – offline generation of a detailed request (prompt, model, pricing) that you must explicitly approve.
• Model routing – supports Google’s Nano Banana 2.1 (default), Nano Banana 2 Lite, Nano Banana Pro and a deprecated 2‑route for 512‑pixel drafts.
• Edit & continue – make a single change while preserving identity, geometry and brand details, or continue a session with stored Nano Banana 2.1/Pro data.
• Compare – run up to three prompts across three model routes in a single bounded portfolio.
• Typeset – add copy, fonts, logos and trusted raster art locally when exact text matters.
• Safety & cost controls – API keys never appear in commands or CI, each paid call shows a plan and requires a one‑time approval, cost logs are private, and raw prompts are omitted by default.

How to get started

1. Install the plugin via the marketplace commands shown in the README.
2. Provide a Gemini API key (billing‑enabled Google AI project) and ensure you are on Linux/macOS (Windows needs WSL).
3. Issue a command like:

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

4. Review the generated plan (prompt, model, estimate) and approve it.
5. The image is created, saved locally, and you can then check, edit, or ship it.

Who it’s for
Designers, marketers, or developers who want a controlled, auditable way to use Google’s Gemini image models without manually crafting prompts or risking unexpected charges. The emphasis on review and approval makes it suitable for production‑grade workflows where cost, privacy and provenance matter.

Limitations

• Works only on Linux/macOS; Windows support is pending (requires WSL).
• Relies on paid Gemini image generation – no free tier.
• Guarantees only probabilistic output; spelling, geometry or policy compliance are not assured.

Where to learn more

• User guide, prompt‑crafting tips and security details are in the docs/ folder.
• The repository includes a CI validation workflow and versioned release notes.

License – MIT.

---

All information above is taken directly from the repository’s README.