A TypeScript/JavaScript port of the popular Python rembg library for background removal in web environments.

This library provides the same functionality as the Python rembg library but designed to run in web browsers using ONNX Runtime Web. It uses the u2net model for background removal.

• Web-native: Runs entirely in the browser using ONNX Runtime Web
• High-quality model: Uses the u2net model for background removal
• Flexible input/output: Supports Uint8Array, PIL-like images, and numpy-like arrays
• Alpha matting: Advanced edge refinement for high-quality cutouts
• Async operations: Non-blocking operations that won't freeze the UI
• TypeScript: Full type safety and excellent IDE support

npm install rembg-ts

Note: Make sure to place your ONNX model file in the public/models/ directory of your web application:

• public/models/u2net.onnx

import { remove, Image } from 'rembg-ts';

// Basic usage with image bytes
async function removeBackground() {
    // Load image from file input or fetch
    const response = await fetch('/path/to/image.jpg');
    const imageBytes = new Uint8Array(await response.arrayBuffer());
    
    // Remove background
    const resultBytes = await remove(imageBytes) as Uint8Array;
    
    // Create blob and display
    const blob = new Blob([resultBytes], { type: 'image/png' });
    const url = URL.createObjectURL(blob);
    
    const img = document.createElement('img');
    img.src = url;
    document.body.appendChild(img);
}

The main function for background removal.

Parameters:

• data: Uint8Array | PILImage | NumpyArray - Input image data
• alpha_matting?: boolean - Enable alpha matting (default: false)
• alpha_matting_foreground_threshold?: number - Foreground threshold (default: 240)
• alpha_matting_background_threshold?: number - Background threshold (default: 10)
• alpha_matting_erode_size?: number - Erosion size (default: 10)
• session?: BaseSession - Custom session object
• only_mask?: boolean - Return only the mask (default: false)
• post_process_mask?: boolean - Apply post-processing to mask (default: false)
• bgcolor?: [number, number, number, number] - Background color (RGBA)
• force_return_bytes?: boolean - Force return as bytes (default: false)

Returns: Promise<Uint8Array | PILImage | NumpyArray>

Create a new session for a specific model.

Parameters:

• modelName: string - Model name ("u2net")

Returns: BaseSession

Create and initialize a session asynchronously.

Parameters:

• modelName: string - Model name ("u2net")

Returns: Promise<BaseSession>

The library includes PIL-like image processing capabilities:

import { Image } from 'rembg-ts';

// Open image from blob
const img = Image.open(blob);

// Resize image
const resized = img.resize([640, 480]);

// Convert image mode
const rgb = img.convert('RGB');

// Save image
const bytes = new io.BytesIO();
img.save(bytes, 'PNG');

import { np } from 'rembg-ts';

// Create arrays
const arr = np.zeros([100, 100, 3]);
const ones = np.ones([50, 50]);

// Array operations
const result = arr.multiply(2).add(ones);

import { remove, new_session_async } from 'rembg-ts';

// Use u2net model
const session = await new_session_async('u2net');
const result = await remove(imageData, false, 240, 10, 10, session);

const result = await remove(
    imageData,
    true,  // Enable alpha matting
    240,   // Foreground threshold
    10,    // Background threshold
    10     // Erosion size
);

const result = await remove(
    imageData,
    false, // alpha_matting
    240,   // alpha_matting_foreground_threshold
    10,    // alpha_matting_background_threshold
    10,    // alpha_matting_erode_size
    undefined, // session
    false, // only_mask
    false, // post_process_mask
    [255, 0, 0, 255] // Red background
);

// Create session once for efficiency
const session = await new_session_async('u2net');

for (const imageData of imageList) {
    const result = await remove(imageData, false, 240, 10, 10, session);
    // Process result...
}

• u2net: General purpose, high quality background removal

• Modern web browser with WebAssembly support
• ONNX Runtime Web
• Model files served from your web server

1. Async operations: All model operations are asynchronous to prevent UI blocking
2. Model loading: Models are fetched from /public/models/ directory
3. Web APIs: Uses Fetch API, Blob, and other web APIs instead of file system operations
4. TypeScript: Full type safety and modern JavaScript features

1. Reuse sessions: Create sessions once and reuse them for multiple images
2. Preload models: Call new_session_async() early to preload models
3. Image sizing: Resize large images before processing for better performance

MIT License - see LICENSE file for details.

Contributions are welcome! Please feel free to submit issues and pull requests.