🍌 nano-banana-cli

A tiny CLI for generating images with Google Gemini — straight from your terminal.

No API key needed. It drives a real browser behind the scenes, signs in with your Google account, and downloads the result.

nanban generate "a golden retriever wearing a tiny wizard hat"

How it works

nanban uses Puppeteer to automate a Chrome/Chromium browser window that talks to Gemini. It navigates to the Gemini image generation UI, types your prompt, waits for the result, and saves the downloaded PNG to disk.

Your Google session is saved in ~/.nban/profile/ so you only need to sign in once.

Prerequisites

• Bun — used to run the TypeScript source directly
• A Google account with access to Gemini

Chrome is auto-detected if installed. If it's not found, Puppeteer's bundled Chromium is used instead.

Installation

The quickest way is to run it directly without installing anything:

# with Bun
bunx nanban generate "a sunset over the ocean"

# with npm
npx nanban generate "a sunset over the ocean"

Or install it globally so nanban is always available:

npm install -g nanban
# or
bun add -g nanban

From source:

git clone https://github.com/hffmnnj/nano-banana-cli
cd nano-banana-cli
bun install
bun run src/index.ts generate "a sunset over the ocean"

Setup: Sign in once

Before generating anything, you'll need to sign in to your Google account:

nanban auth

This opens a real browser window. Sign in normally, then close the window when you're done. Your session is saved and reused for future generations — you won't need to do this again unless the session expires.

Generating images

Single image

nanban generate "a rainy Tokyo street at night, neon reflections"

The image is saved as nban-{timestamp}.png in your current directory.

Custom output path

nanban generate "a minimalist mountain landscape" --output ./art/mountain.png

Multiple images at once

nanban generate "an astronaut tending a garden on mars" --count 4

Each image gets its own file: mountain-1.png, mountain-2.png, etc. They're generated in parallel across browser tabs, so it's as fast as Gemini allows.

All options

nanban generate <prompt> [options]

Arguments:
  prompt              Your image generation prompt (required)

Options:
  -o, --output        Output file path (default: ./nban-{timestamp}.png)
  -n, --count         Number of images to generate (default: 1)
  -v, --verbose       Print detailed automation logs
      --headed        Run the browser visibly (useful for watching what's happening)
  -d, --debug         Write a debug log to ~/.nban/debug/ and keep the browser open on failure

Debugging

If something isn't working, --headed lets you watch the browser in real time:

nanban generate "a stormy ocean" --headed

For deeper inspection, --debug writes a timestamped log to ~/.nban/debug/:

nanban generate "a stormy ocean" --headed --debug

When --debug is active, the browser stays open on failure so you can see exactly where things went wrong.

File locations

| Path | Purpose | |------|---------| | ~/.nban/profile/ | Saved browser session (your Google login) | | ~/.nban/debug/ | Debug logs (only written with --debug) | | ./nban-{timestamp}.png | Default output location |

Notes & limitations

• This tool automates the Gemini web interface. It's not using an official API, so it may break if Google changes the UI.
• Image generation can take 30–60+ seconds depending on Gemini's load.
• A 6-minute generation watchdog is in place — if Gemini takes longer than that, the request is abandoned and an error is returned.
• If your session expires, just run nanban auth again.
• Gemini's image generation requires the Pro model. nanban switches to it automatically if Gemini defaults to Fast.

License

MIT — see LICENSE.

Made by James Hoffmann.