agentleFS
ckorhonenckorhonen/claude-skills/skills/nano-banana/SKILL.md

nano-banana

Skill17 starsChanged 4 months agoMITView on GitHub

Generate and edit high-quality AI images using Google's Gemini image models (Nano Banana series) via MCP. Use when user wants to create images, edit photos, generate graphics, or needs visual content with text rendering. Supports gemini-3.1-flash-image-preview (Nano Banana 2) and gemini-2.5-flash-image-preview (Nano Banana).

  • Reads credentials

40 other projects use a version of this, 1 of them edited.

Adopted by

Most recently taken up by

Used with

Projects with it also use agency-docs-updater, brand-agency and 10 more.

Other files people use alongside this one.

Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
  • KISHOR403Skill

    agency-docs-updater

    End-to-end pipeline for publishing Claude Code lab meetings. Automatically finds/creates Fathom transcript, downloads video, uploads to YouTube, generates fact-checked Russian summary, creates MDX documentation, and pushes to agency-docs for Vercel deployment. Single invocation replaces 5+ manual steps.

    Also in 3 projects that use this

  • KISHOR403Skill

    brand-agency

    Applies Agency brand colors and typography to artifacts including presentations, SVG graphics, documents, and web interfaces. This skill should be used when brand colors, visual formatting, neobrutalism style, or Agency design standards apply. Keywords - branding, corporate identity, visual identity, styling, brand colors, typography, visual formatting, visual design, neobrutalism.

    Also in 3 projects that use this

  • jechearteSkill

    elevenlabs-tts

    Generate realistic audio from text using ElevenLabs Text-to-Speech API. Use when the user needs to convert text to speech, create voiceovers, generate narration, or produce audio content. Triggers include "generate audio", "text to speech", "TTS", "voiceover", "narration", "ElevenLabs", "audio from text", "read this text aloud

    Also in 5 projects that use this

  • +9 more from KISHOR403

The file

We split this skill into the rules it gives an agent

5 rules. 0 are common, 3 are distinctive, and 2 are very specific to this project.

  • 0

    Common

    Most of the 36 edited versions we read keep these.

  • 3

    Distinctive

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Use history:0 reference in subsequent prompts

    line 197
  2. Distinctive:

    Never committed to version control

    line 244
  3. Very specific:

    Use .env files for key management in projects

    line 246
  4. Distinctive:

    Process Mapper: - Create diagram visuals

    line 272
  5. Distinctive:

    Research to Essay: - Add illustrations to content

    line 273

“Common” is measured: kept by at least 6 in 10 of the 36 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Nano Banana - AI Image Generation via MCP

Generate stunning 4K images, edit photos, and create graphics with perfect text rendering using Google's Gemini image models via MCP. The default model is gemini-3.1-flash-image-preview (Nano Banana 2), launched February 2026.

When to Use

Invoke when user:

  • Asks to "generate an image" or "create a picture"
  • Wants to "edit this photo" or "modify this image"
  • Needs graphics with text (logos, infographics, diagrams)
  • Requests "consistent characters" across multiple images
  • Says "visualize this" or "make me a [visual thing]"

Prerequisites

1. Gemini API Key

Get a free API key from Google AI Studio:

  1. Sign in with Google account
  2. Click "Get API Key" → "Create API Key"
  3. Copy and save securely

2. MCP Server Setup

Recommended: NanoBanana-MCP (uses Gemini 3.1 Flash Image for highest quality)

# Quick install via Claude Code CLI
claude mcp add nano-banana --env GEMINI_API_KEY=your-key-here -- npx -y nanobanana-mcp

Or add to ~/.claude/settings.json manually:

{
  "mcpServers": {
    "nano-banana": {
      "command": "npx",
      "args": ["-y", "nanobanana-mcp"],
      "env": {
        "GEMINI_API_KEY": "your-api-key-here"
      }
    }
  }
}

Alternative: Nano-Banana-MCP by ConechoAI (Gemini 2.0 Flash - faster, lower cost)

{
  "mcpServers": {
    "nano-banana": {
      "command": "npx",
      "args": ["nano-banana-mcp"],
      "env": {
        "GEMINI_API_KEY": "your-api-key-here"
      }
    }
  }
}

Available Tools

Once MCP is configured, these tools become available:

Core Tools

ToolPurposeKey Parameters
gemini_generate_imageCreate new images from text promptsprompt, model, aspectRatio, imageSize
gemini_edit_imageModify existing images with instructionsimagePath, instructions, model
continue_editingRefine the last generated imageinstructions
get_image_historyList all generated images in session-

Model Options

Model IDDescription
gemini-3.1-flash-image-previewDefault. Highest quality, 4K support, best text rendering (Nano Banana 2, launched Feb 2026)
gemini-2.5-flash-image-previewFast generation, 1K resolution, good quality (Nano Banana)
gemini-3-pro-image-previewPrevious-gen Pro model, still high quality (Nano Banana Pro)
gemini-2.0-flash-expOlder experimental model; prefer gemini-2.5-flash-image-preview for new work

Image Size (Gemini 3.1 / 3 only)

SizeUse Case
4KFinal assets, print, marketing materials
2KBalanced quality and speed
1KFast iteration, prototyping

Advanced Features

FeatureCapability
4K OutputUp to 5632×3072 pixels
Text RenderingAccurate text in images (signs, labels, UI)
Multi-Image CompositionCombine up to 14 reference images
Character ConsistencyMaintain same character across 5+ images
Google Search GroundingReal-world accurate imagery

Prompting Best Practices

Structure Your Prompts

[Subject] + [Style] + [Details] + [Technical Specs]

Example:

"A cozy coffee shop interior, watercolor illustration style, warm lighting, wooden furniture, steaming cup on table, 4K resolution, soft morning light through windows"

For Best Results

  1. Be Specific - Include colors, materials, lighting, mood
  2. Specify Style - "photorealistic", "oil painting", "3D render", "anime"
  3. Add Context - Time of day, weather, setting
  4. Request Resolution - "4K", "high resolution", "detailed"

Precision Mode (JSON Prompting)

For high-stakes work requiring exact reproducibility, use structured JSON schemas.

When to Activate

Trigger phrases:

  • "I need exact control over..."
  • "Create a product shot for [brand]..."
  • "Generate a UI mockup..."
  • "Make an infographic showing..."
  • "I want to iterate on just the lighting..."
  • "A/B test different versions..."

Three Schema Types

TypeUse CaseKey Controls
marketing_imageProduct shots, hero imagessubject, props, lighting, camera, brand locks
ui_builderApp screens, dashboardstokens, screens, containers, components
diagram_specFlowcharts, infographicsnodes, edges, data constraints

The Translator Workflow

  1. Describe - User explains what they want in plain English
  2. Clarify - Claude asks targeted questions for missing fields
  3. Generate - Claude outputs structured JSON schema
  4. Review - User checks key fields match intent
  5. Render - JSON converts to precise prompt for Nano Banana 2
  6. Iterate - Modify specific fields, re-render (scoped changes)

Example: Product Shot

User: "I need a hero shot for Aurora Lime seltzer"

Claude asks: "For the Aurora Lime hero shot:

  1. Can size? (12oz standard?)
  2. Props? (lime slices, ice, condensation?)
  3. Background style? (solid color, gradient, bokeh?)
  4. Lighting mood? (bright/refreshing or moody/premium?)"

Result: Structured JSON with exact specifications that can be iterated field-by-field.

Scoped Edits (The Key Unlock)

JSON enables changing ONE thing without regenerating everything:

ChangeWhat Stays Fixed
Swap lighting directionSubject, props, background
Try different camera angleLighting, props, environment
Change background colorSubject geometry, lighting setup
Add/remove propsEverything else

Reference Docs

  • references/json-prompting.md - Full JSON prompting guide
  • references/translator-prompt.md - Translator system prompt
  • references/schemas/ - Template schemas for each type
  • references/examples-json.md - Filled-out examples

Text in Images

Nano Banana 2 excels at text rendering:

"A vintage movie poster for 'COSMIC ADVENTURE' with bold retro typography, starfield background, astronaut silhouette, 1970s sci-fi aesthetic"

Character Consistency

For consistent characters across images:

  1. Generate initial character with detailed description
  2. Use history:0 reference in subsequent prompts
  3. Describe scene changes while referencing original
First: "A young woman with red curly hair, freckles, green eyes, wearing a blue jacket"
Then: "The same woman from history:0, now sitting at a café, reading a book"

Workflow Examples

Basic Image Generation

User: "Create an image of a futuristic city at sunset"

Claude uses: gemini_generate_image
Prompt: "Futuristic cityscape at golden hour sunset, towering glass skyscrapers with holographic advertisements, flying vehicles, warm orange and purple sky, photorealistic, 4K resolution, cinematic lighting"

Photo Editing

User: "Edit this photo to make it look like winter"

Claude uses: gemini_edit_image
Input: [user's image path]
Instructions: "Transform to winter scene: add snow on ground and surfaces, frost on windows, visible breath, overcast sky, cool blue color grading"

Iterative Refinement

User: "Make the lighting warmer"

Claude uses: continue_editing
Instructions: "Adjust lighting to warmer tones, add golden hour glow, enhance orange/yellow highlights, softer shadows"

Output Management

Images save to: ~/Documents/nanobanana_generated/

Naming format: generated-[timestamp]-[id].png

Security Notes

  • API keys stored locally in environment variables
  • Never committed to version control
  • Images processed locally, not stored on external servers
  • Use .env files for key management in projects

Model Comparison

ModelSpeedQualityMax ResCostBest For
gemini-3.1-flash-image-previewFastHighest4K~$0.045/img (1K)Final assets, print, marketing
gemini-2.5-flash-image-previewFastestGood1K~$0.039/imgPrototyping, iteration, drafts
gemini-3-pro-image-previewSlowerHigh4K~$0.134/imgPrevious-gen Pro fallback
gemini-2.0-flash-expFastGood1KLowLegacy; use 2.5-flash-image-preview instead

Pricing as of March 2026. Free tier (~500 req/day) available via Google AI Studio for all models.

Troubleshooting

IssueSolution
"API key invalid"Verify key at AI Studio
"Rate limited"Wait 60s, or upgrade API tier
"MCP not connected"Restart Claude Code, check config syntax
"Image not saving"Check write permissions on output directory

Integration

Works well with:

  • Artifacts Builder - Generate images for HTML artifacts
  • Process Mapper - Create diagram visuals
  • Research to Essay - Add illustrations to content

References

  • references/prompting-guide.md - Detailed prompting techniques
  • references/examples.md - Sample prompts by category

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.