Nano Banana images with Gemini CLI

Overview

Use Gemini CLI plus the Nano Banana extension to generate and edit images via slash commands. Outputs are saved to ./nanobanana-output/ for review or delivery.

Prerequisites

• Confirm Gemini CLI is installed: gemini --version
• Confirm the extension is installed: gemini extensions install https://github.com/gemini-cli-extensions/nanobanana
• Ensure an API key is set: NANOBANANA_GEMINI_API_KEY, NANOBANANA_GOOGLE_API_KEY, GEMINI_API_KEY, or GOOGLE_API_KEY
• Optional: set NANOBANANA_MODEL=gemini-3-pro-image-preview for Nano Banana Pro
• If MCP tools are blocked by policy, run with approvals: gemini --approval-mode yolo -p "..." or allow the MCP server with --allowed-mcp-server-names nanobanana

Workflow

1. Identify the request type (generate, edit, restore, icon, pattern, story, diagram).
2. Ensure any input images are in the current directory or the extension search paths.
3. Start Gemini CLI in the working directory: gemini (or use gemini --approval-mode yolo -p "..." for one-shot runs)
4. Run the appropriate command:
   • /generate "prompt" --count=3 --styles="watercolor,sketch"
   • /edit input.png "instruction" --preview
   • /restore old_photo.jpg "instruction"
   • /icon "prompt" --sizes="64,128,256"
   • /pattern "prompt" --type="seamless"
   • /story "prompt" --steps=4
   • /diagram "prompt" --type="architecture"
   • /nanobanana "freeform instruction"
5. Collect outputs from ./nanobanana-output/ and share the file paths.

Troubleshooting

• If generate_image is denied by policy, re-run with --approval-mode yolo or allow MCP server nanobanana.
• If you see No valid API key found, set NANOBANANA_GEMINI_API_KEY (or NANOBANANA_GOOGLE_API_KEY, GEMINI_API_KEY, GOOGLE_API_KEY) and retry.

Prompting Guidance

• Specify subject, style, medium, lighting, composition, color palette, and constraints.
• For consistent series, repeat style and include specific variation goals.
• Use --count, --styles, and --variations for systematic exploration.
• Use --preview only when interactive output is desired.

Files and Tools

• Use scripts/example.py to list output images or fetch the newest output.
• Use references/api_reference.md for command options and file-handling details.
• Use assets/example_asset.txt as a prompt template.