nano-banana-ultimate

Nano Banana Ultimate

The best image generation plugin for Claude Code. Generate and edit images via Google Gemini with domain-specialized auto-prompt building. You describe what you want in plain language — the plugin builds an optimized prompt following Google's official best practices.

Requirements

• uv installed
• GEMINI_API_KEY environment variable set — get one here

Installation

/plugin install nano-banana-ultimate@devx-plugins

Skills

Skill	Trigger	Default Ratio	Description
banana	"generate image", "edit image", "draw"	1:1	General-purpose catch-all for any image
banana-ads	"ad", "banner", "social ad"	varies by format	Online advertising visuals with format-aware sizing
banana-web	"hero background", "landing page", "web design"	16:9	Hero backgrounds, card images, section dividers
banana-youtube	"thumbnail", "youtube"	16:9	Bold, click-optimized YouTube thumbnails
banana-illustration	"illustration", "watercolor", "pixel art"	1:1 / 4:3	Flat vector, watercolor, 3D clay, pixel art, isometric
banana-logo	"logo", "brand mark", "monogram"	1:1	Minimal logo concepts and brand identity marks
banana-people	"portrait", "headshot", "lifestyle photo"	3:4 / 3:2	Portraits, editorial shots, team photos
banana-objects	"product photo", "still life", "food photo"	1:1 / 4:5	Product photography, mockups, food shots

Quick Start

/banana-youtube coding tutorial about async/await

/banana-logo minimal tech startup called "Nexus" in blue tones

/banana-ads Instagram story ad for a meditation app

Or just say what you want — Claude routes to the right skill automatically.

Models

Model	Name	Speed	Quality	Best For
gemini-3-pro-image-preview	Nano Banana Pro	Slower	Highest	Final output, complex scenes
gemini-3.1-flash-image-preview	Nano Banana 2	Fast	Good	Drafts, iterations, simple images

Default: Nano Banana Pro for maximum quality.

Features

• Auto-prompt building — You say "a product photo of headphones", Claude builds a detailed prompt with lighting, camera specs, composition, and materiality
• 8 domain-specialized skills — Each skill applies domain-specific defaults (aspect ratio, resolution, lighting style, composition rules)
• Image editing — Pass existing images with --input-image for background swaps, style transfers, and refinements (up to 14 reference images)
• Google best practices — Every prompt uses positive framing, strong verb openers, camera/lens specs, and materiality descriptions
• 63 reference prompts — Curated examples across all domains for inspiration and consistency
• Resolution control — 1K for drafts, 2K for production, 4K for print
• 10 aspect ratios — From 1:1 squares to 21:9 ultrawide banners

How It Works

1. You describe what you want in plain language
2. Claude identifies the domain and applies the right skill
3. The skill builds an optimized Gemini prompt using the formula: [Subject] + [Action] + [Location] + [Composition] + [Style]
4. A Python script calls the Google Gemini API and saves a PNG
5. Claude reports the file path and offers to iterate

File Structure

nano-banana-ultimate/
├── .claude-plugin/plugin.json        # Plugin manifest
├── scripts/banana.py                 # Single Python script (google-genai + Pillow)
├── skills/
│   ├── banana/                       # General-purpose catch-all
│   ├── banana-ads/                   # Online advertising
│   ├── banana-web/                   # Web design assets
│   ├── banana-youtube/               # YouTube thumbnails
│   ├── banana-illustration/          # Illustrations (7 sub-styles)
│   ├── banana-logo/                  # Logos and brand identity
│   ├── banana-people/                # People and portraits
│   └── banana-objects/               # Products and objects
└── README.md