nano-banana

name	nano-banana
effort	low
description	Generate images using Google's Nano Banana 2 model. Invoke with /nano-banana followed by a prompt description, or when the user asks to generate, create, or make an image.

Nano Banana — Image Generation Skill

Generate images using Google's Nano Banana 2 (Gemini 3.1 Flash Image) model.

Project Config Detection

On every invocation, check for .nano-banana.json in the current working directory.

• If found: Load the config and apply it during prompt enhancement (see "Using the Medium Config" below).
• If not found: Ask the user — "No .nano-banana.json found in this project. Want me to set one up, or generate without a style profile?" If they want setup, follow the "Setup Flow" below.

Setup Flow

When the user wants to create a .nano-banana.json for their project, interview them with these questions:

1. What kind of content is this for? (personal blog, company blog, food blog, product marketing, social media, documentation, etc.)
2. Describe the visual feel you're going for. Offer starter examples if they're unsure:
   • Personal blog: "Warm, authentic, editorial. Natural lighting, shallow depth of field, Fujifilm color science."
   • Company blog: "Clean, professional, modern. Studio softbox lighting, crisp focus, minimal backgrounds."
   • Food blog: "Inviting overhead or 45° shots. Natural window light, rustic surfaces, rich warm tones."
   • Product marketing: "Polished, high-contrast studio shots. Dramatic lighting, clean backgrounds, premium feel."
   • Social media: "Bold, eye-catching, vibrant saturation. Strong composition with room for text overlay."
3. Default aspect ratio? Suggest based on their medium (blog hero → 16:9, social → 1:1, Pinterest → 9:16, etc.)
4. Default resolution? (0.5K, 1K, 2K, 4K — suggest 2K as a good default)
5. Any recurring visual elements? Things that should appear consistently (brand colors, specific props, surface textures, typography style).
6. Camera and lens preferences? (e.g., "shot on Fujifilm" for warm tones, "macro lens" for detail, "GoPro" for action)

After the interview, generate the .nano-banana.json and write it to the project root. The schema:

{
  "medium": "<medium name>",
  "defaults": {
    "aspect_ratio": "<ratio>",
    "resolution": "<resolution>"
  },
  "style": {
    "description": "<one-line summary of the visual identity>",
    "lighting": "<lighting direction>",
    "color_grading": "<color/film stock direction>",
    "composition": "<default composition approach>",
    "camera": "<camera and lens defaults>",
    "materiality": "<textures, surfaces, materials>"
  },
  "recurring_elements": [
    "<element 1>",
    "<element 2>"
  ]
}

See config.example.json in the skill directory for a complete example.

Usage

The user provides an image description. Your job is to:

1. Enhance the prompt using the Nano Banana prompting framework, applying the medium config if available (see below).

2. Show the user the enhanced prompt so they can approve or tweak it before generating.

3. Run the generation script using the Bash tool:

   uv run ~/.claude/skills/nano-banana/generate.py "the enhanced prompt" -o output-filename.png
   

   Optional flags (override config defaults):

   • -i / --input: input image for editing/refinement (enables image-to-image mode)
   • -r / --resolution: 0.5K, 1K, 2K (default), 4K
   • -a / --aspect-ratio: 1:1, 2:3, 3:2, 3:4, 4:3, 4:5, 5:4, 9:16, 16:9, 21:9, etc.
   • -o / --output: output file path (default: generated-image.png)
   • -c / --config: path to config file (default: ./.nano-banana.json)

   When a .nano-banana.json exists, the script reads defaults from it automatically. CLI flags override config values.

4. Show the result to the user by reading the generated image file with the Read tool.

5. Iterate — when the user wants changes, ask: "Refine the existing image, or start from scratch?"

   • Refine existing: Pass the previously generated image as input with -i so the model uses it as a reference. Adjust the prompt to describe only what should change (e.g., "Make the background warmer" or "Add a coffee cup to the left side").

     uv run ~/.claude/skills/nano-banana/generate.py "Make the background warmer and add soft bokeh" -i previous-output.png -o refined-output.png
     

   • : Run without , using a revised prompt from the ground up.

Using the Medium Config for Prompt Enhancement

When a .nano-banana.json is loaded, weave its style properties into the [Style] segment of the prompting framework:

Formula: [Subject] + [Action] + [Location/context] + [Composition] + [Style]

How each config field maps to the prompt:

Config field	Prompt segment	Example
style.composition	[Composition]	"Overhead flat-lay"
style.lighting	[Style] lighting clause	"Soft natural window light, diffused"
style.color_grading	[Style] color clause	"Warm, rich tones"
style.camera	[Style] camera clause	"Shot on Fujifilm X-T5, 56mm f/1.2"
style.materiality	[Location/context] surfaces	"On rustic wood table with linen"
recurring_elements	Append to scene description	"Fresh herbs as garnish props"

The user's specific request always takes priority — the config provides defaults that fill in gaps the user didn't specify.

Prompting Tips (from the Nano Banana guide)

• Be specific: concrete details on subject, lighting, composition
• Use positive framing: describe what you want, not what you don't want (e.g. "empty street" instead of "no cars")
• Control the camera: use photographic terms like "low angle", "aerial view", "f/1.8"
• Define lighting: "three-point softbox", "golden hour backlighting", "chiaroscuro"
• Specify materiality: textures, fabrics, surfaces
• Choose film stock/color grading: "1980s color film", "muted teal tones"
• Use quotes for any text that should appear in the image
• Name specific cameras for visual DNA: "shot on Fujifilm", "GoPro fisheye"

Requirements

• GOOGLE_API_KEY environment variable must be set
• Uses uv run with inline dependencies (no pre-install needed)