🎨 GenMix

AI-powered image generator supporting Google Gemini and Fal Nano Banana 2. Supports image generation from text prompts and image modification with reference images (Gemini).

Features ✨

• 🖼️ Image generation from text descriptions
• 🎨 Image modification using reference images
• 🔄 Style transfer - Apply artistic styles to your images
• 📐 Quality control - Generate in 1K, 2K or 4K
• 🎯 Multiple formats - Supports local paths, URLs and Base64
• 💾 Auto-save with unique hash per prompt

Installation

npm install genmix

AI Skill

You can also add GenMix as a skill for AI agentic development:

npx skills add https://github.com/clasen/GenMix --skill genmix

Setup

Create a .env file in your project root:

GEMINI_API_KEY=your_api_key_here
FAL_API_KEY=your_fal_api_key_here

GEMINI_API_KEY is used with provider gemini and FAL_API_KEY is used with provider fal.

Basic Usage

The Power of GenMix: Multiple References & Chainable API

GenMix shines when combining multiple images with specific instructions using its intuitive chainable API:

import { GeminiGenerator } from 'genmix';
const generator = new GeminiGenerator();

const result = await generator
  .pro() // Use the Pro model for best results
  .addReference('./person.jpg', 'Use this person as the main subject')
  .addReference('./background.jpg', 'Use this as the background setting')
  .generate('A photo of the person standing in the background setting, cinematic lighting');

await generator.save({ filename: 'composite-result' });

Provider Selection (Gemini or Fal)

import { GeminiGenerator, FalGenerator } from 'genmix';

const gemini = new GeminiGenerator({ apiKey: process.env.GEMINI_API_KEY });
const fal = new FalGenerator({ apiKey: process.env.FAL_API_KEY });

await gemini.flash().generate('A cinematic portrait with dramatic lighting');
await fal.generate('A cinematic portrait with dramatic lighting', {
  numberOfImages: 1,
  quality: '1K',
  aspectRatio: '1:1'
});

Fal models available in this integration:

• flash (or banana2) -> fal-ai/nano-banana-2/edit (image-to-image editing)
• pro (or banana-pro) -> fal-ai/nano-banana-pro/edit (image-to-image editing)

Model Selection

You can easily switch between the Pro and Flash models using chainable methods:

import { GeminiGenerator } from 'genmix';

const generator = new GeminiGenerator();

// Use the Pro model
await generator
  .pro()
  .generate('A highly detailed portrait of a cat');

// Switch to the Flash model
await generator
  .flash()
  .generate('A quick sketch of a dog');

Simple Image Generation

import { GeminiGenerator } from 'genmix';

const generator = new GeminiGenerator();

// Generate an image
const result = await generator.generate(
  'A futuristic city with flying cars, cyberpunk style',
  {
    numberOfImages: 1,
    quality: '2K',
    aspectRatio: '16:9'
  }
);

// Save images
const savedPaths = await generator.save({ directory: './output' });
console.log('Images saved:', savedPaths);

Image Modification with Reference Images

// Modify an existing image
const result = await generator.generate(
  'Transform this image to have sunset lighting with warm orange tones',
  {
    referenceImage: './my-image.png',  // Local path
    quality: '2K',
    numberOfImages: 1
  }
);

const savedPaths = await generator.save({ directory: './output' });

Using URLs as Reference

const result = await generator.generate(
  'Convert this photo into a watercolor painting',
  {
    referenceImage: 'https://example.com/image.jpg',  // URL
    quality: '1K'
  }
);

generator.save({ directory: './output' });

Configuration Options

Constructor

new GeminiGenerator({
  apiKey: string,           // Your Google API key (required)
  modelId: string           // Model to use (optional, default: 'gemini-3-pro-image-preview')
})

new FalGenerator({
  apiKey: string,           // Your Fal API key (required)
  modelId: string           // Optional: FalGenerator.MODELS.BANANA_2 or BANANA_PRO_EDIT
})

Model Selection Methods

generator.pro()    // Switches to the gemini-3-pro-image-preview model
generator.flash()  // Switches to the gemini-3.1-flash-image-preview model

Both methods are chainable and return the generator instance.

Fal generator model methods:

fal.banana2()   // fal-ai/nano-banana-2/edit (image editing)
fal.bananaPro() // fal-ai/nano-banana-pro/edit (image editing)
fal.pro()       // alias of bananaPro()
fal.flash()     // alias of banana2()

Reference Methods

You can also use chainable methods to add one or multiple reference images before calling generate():

generator.addReference(image, description) // Adds a reference image (path, URL, Buffer)
generator.clearReferences()                // Removes all queued reference images

For fal (flash and pro), references can be URL, data URI, local file path, or Buffer.

generate() Method

await generator.generate(prompt, options)

Parameters:

| Option | Type | Description | Default | | ------------------------ | ------------- | ------------------------------------------- | ------- | | prompt | string | Description of what you want to generate | - | | options.referenceImage | string/Buffer | Reference image (path, URL, Base64, Buffer) | - | | options.numberOfImages | number | Number of images to generate | 1 | | options.quality | string | Quality: '1K', '2K', '4K' | - | | options.aspectRatio | string | Aspect ratio: '1:1', '16:9', '4:3', etc. | - | | options.width | number | Final output width in pixels (requires height) | - | | options.height | number | Final output height in pixels (requires width) | - |

save() Method

await generator.save(options)

Parameters:

| Option | Type | Description | Default | | ------------------- | ------ | ------------------------------------------------ | ------- | | options.directory | string | Target directory path to save images | '.' | | options.filename | string | Custom filename (without extension). If not provided, uses hash-based filename | - | | options.extension | string | File format: 'jpg', 'png', 'webp', 'avif', 'tiff' | 'jpg' |

Examples:

// Save to specific directory with auto-generated filename (jpg by default)
await generator.save({ directory: './output' });

// Save as PNG
await generator.save({ directory: './output', extension: 'png' });

// Save to specific directory with custom filename
await generator.save({ directory: './output', filename: 'my-image' });

// Save as WebP with custom filename
await generator.save({ directory: './output', filename: 'my-image', extension: 'webp' });

// Save to current directory with custom filename
await generator.save({ filename: 'my-image' });

// Save to current directory with auto-generated filename (jpg)
await generator.save();

Smart target size in library mode (non-CLI)

You can request high generation quality and still force an exact final output size directly in generate():

const generator = new GeminiGenerator();

await generator.generate('App icon, flat minimal style', {
  quality: '4K', // generation quality (independent)
  width: 400,
  height: 400
});

// width/height resize is automatically applied on save()
await generator.save({ filename: 'icon-400x400', extension: 'png' });

If you also pass aspectRatio, it must match the ratio derived from width/height.

Note:

• When multiple images are generated and a custom filename is provided, they will be saved as filename_0.jpg, filename_1.jpg, etc.
• The method uses Sharp for image conversion, supporting high-quality format conversion

Advanced Examples

Style Transfer

const result = await generator.generate(
  'Transform this photo into a Van Gogh style painting with visible brush strokes',
  {
    referenceImage: './photo.jpg',
    quality: '4K'
  }
);

Lighting Modification

const result = await generator.generate(
  'Change the lighting to dramatic studio lighting with strong shadows',
  {
    referenceImage: './portrait.png',
    quality: '2K'
  }
);

Multiple Variations

const result = await generator.generate(
  'Add dramatic clouds and enhance colors',
  {
    referenceImage: './landscape.jpg',
    numberOfImages: 3,
    quality: '1K'
  }
);

// Generates 3 variations of the same modification

Using Buffers

import fs from 'fs';
const imageBuffer = fs.readFileSync('./image.png');

const result = await generator.generate(
  'Make this image look cinematic',
  {
    referenceImage: imageBuffer
  }
);

Reference Image Formats

GenMix accepts reference images in multiple formats:

1. Local file path: './image.png'
2. URL: 'https://example.com/image.jpg'
3. Data URI: 'data:image/png;base64,iVBORw0KG...'
4. Buffer: Buffer.from(...)

Supported image formats: PNG, JPEG, GIF, WEBP

Error Handling

try {
  const result = await generator.generate(prompt, options);
  
  if (result.images && result.images.length > 0) {
    const paths = generator.save({ directory: './output' });
    console.log('Success!', paths);
  } else {
    console.log('No images generated');
  }
} catch (error) {
  console.error('Error:', error.message);
  
  // Common errors:
  // - 'API Key is required'
  // - 'Failed to read reference image file'
  // - 'Failed to download reference image from URL'
  // - 'Gemini API Error: ...'
}

Project Structure

genmix/
└── generators/
│   ├── BaseGenerator.js      # Base class with utilities
│   ├── GeminiGenerator.js    # Gemini API implementation
│   └── FalGenerator.js       # Fal Nano Banana 2 implementation
├── demo/
│   ├── example.js                # Basic examples
│   └── example-translation.js    # Translate image
├── index.js                      # Entry point
└── README.md

Best Practices

1. Clear Prompts: Be specific about what you want

   // ✅ Good
   'Add dramatic sunset lighting with orange and pink tones in the sky'
      
   // ❌ Vague
   'Make it better'

2. Appropriate Quality:

   • 1K: Quick tests
   • 2K: General use
   • 4K: High quality (slower)

3. Image Size: Reference images between 512x512 and 2048x2048 work best

4. Result Caching: Images are automatically saved with unique hash based on the prompt

CLI Usage

Global CLI Installation

Install GenMix globally to use it from the command line:

npm install -g genmix

First run and API key persistence

If no API key is available, the CLI asks for it on first use and saves it to:

~/.genmix/config.json

You can set or update it explicitly anytime:

genmix --config

API key resolution order in CLI:

1. Provider-specific environment variable (GEMINI_API_KEY or FAL_API_KEY)
2. Saved config (~/.genmix/config.json) using geminiApiKey/apiKey or falApiKey
3. Interactive prompt (then persisted)

Basic CLI commands

# Show help
genmix --help

# Generate from prompt
genmix "A futuristic city with flying cars, cyberpunk style"

# Generate multiple images
genmix "A cozy cabin in winter" -n 2 -q 2K -r 16:9 -m flash

# Use Fal Nano Banana 2 edit (flash) with reference
genmix "Restyle this room with warm sunset mood" --provider fal -m flash --ref "./room.jpg" -n 2 -q 2K -r 16:9

# Use Fal Nano Banana Pro (edit) with reference
genmix "make this scene cinematic" --provider fal -m banana-pro --ref "https://example.com/input.png"

Output file path or directory

--output accepts either:

• a directory path, or
• a full output file path (including filename + extension)

# Save to a directory (auto-generated hash filename)
genmix "Watercolor fox logo" --output ./output

# Save to exact file path and filename
genmix "Watercolor fox logo" --output ./output/logo-fox.png

Smart target size (independent from generation quality)

You can ask the model for high generation quality (for example 4K) and still force a final exact output size.

When you pass target dimensions, GenMix CLI:

1. Derives the generation ratio automatically (for example 400x400 -> 1:1, 1920x1080 -> 16:9)
2. Generates using your selected quality (1K, 2K, or 4K)
3. Resizes the final image to the exact dimensions you requested

# Ask for 4K quality, deliver exact 400x400 output
genmix "app icon, flat minimal style" -q 4K --width 400 --height 400 --output ./output/icon.png

If you also pass --ratio, it must match the derived ratio from the target size.

References with optional text description

Use --ref <path:text> to add reference images with optional guidance text:

# Reference path only
genmix "Restyle this room" --ref ./room.jpg

# Reference path + description
genmix "Restyle this room" --ref "./room.jpg:keep composition and camera angle"

# Multiple references with descriptions
genmix "Create product ad scene" \
  --ref "./product.png:use as main subject" \
  --ref "./bg.jpg:use as background mood"

# References for fal models (URL, data URI, or local path)
genmix "Edit this image for a magazine look" \
  --provider fal -m flash \
  --ref "./photo.png"

CLI options

-n, --number <N>          Number of images (default: 1)
-q, --quality <1K|2K|4K>  Image quality (default: 1K)
-p, --provider <gemini|fal> Provider (default: gemini)
-r, --ratio <ratio>       Aspect ratio (default: 1:1 for gemini, auto for fal)
-m, --model <...>         gemini: pro|flash (default: flash)
                          fal: pro|flash (aliases: banana-pro|banana2|2, default: flash)
-o, --output <path>       Output directory or full output file path
-f, --format <format>     Output format when output is a directory (default: jpg)
--width <px>              Final output width in pixels (requires --height)
--height <px>             Final output height in pixels (requires --width)
--ref <path[:text]>       Reference image (path/URL/data URI); for URL descriptions use URL::description
--no-sharp                Save raw model bytes without Sharp conversion (disables resizing)
--config                  Set/update persisted API key
--help                    Show help

Additional Resources

• Code Examples
• Google Gemini API Documentation
• Fal Nano Banana 2 Edit Documentation
• Fal Nano Banana Pro Edit Documentation

License

MIT

Contributing

Contributions are welcome! Please open an issue or pull request.