AI Background Remover

A high-performance AI background removal API built with multiple state-of-the-art models. Remove backgrounds from images with professional quality results in seconds.

Quick Start

# Basic usage - transparent background
result = predict(image="your-image.jpg")

# High quality with custom background
result = predict(
    image="portrait.jpg",
    model="isnet-general-use", 
    resize_for_speed=False
)

Features

Multiple AI Models

• u2net - Fast, general purpose (default)
• u2net_human_seg - Optimized for people and portraits
• isnet-general-use - Highest quality, best edges

Performance Optimized

• Smart Resizing - Automatically resizes large images for speed
• Model Caching - Models load once and stay in memory
• CPU Optimized - Fast processing without GPU requirements

Quality Options

• Alpha Matting - Professional edge quality (on high-quality model)
• Multiple Sizes - Works with images from 300x300 to 4000x4000 pixels
• Format Support - PNG, JPG, JPEG, WebP, AVIF input

Parameters

Required

• image: Input image file

Optional

• model: AI model to use
• "u2net" (default) - Fast and good quality
• "u2net_human_seg" - Best for people/portraits

• "isnet-general-use" - Highest quality with alpha matting

• resize_for_speed: Resize large images for faster processing

• true (default) - Resize images >1024px for speed
• false - Keep original size for maximum quality

Use Cases & Recommendations

E-commerce & Product Photos

# Fast processing for product catalogs
result = predict(
    image="product.jpg",
    model="u2net",
    resize_for_speed=True
)

Portrait & People Photos

# Best quality for portraits
result = predict(
    image="portrait.jpg", 
    model="u2net_human_seg",
    resize_for_speed=False
)

High-End Photography

# Maximum quality with alpha matting
result = predict(
    image="professional_photo.jpg",
    model="isnet-general-use", 
    resize_for_speed=False
)

Performance Comparison

Technical Details

• Processing Time: 1-3 seconds per image
• Memory Usage: ~2GB RAM
• Input Formats: PNG, JPG, JPEG, WebP, AVIF
• Output Format: PNG with transparency
• Max Resolution: 4000x4000 pixels recommended

Tips for Best Results

Input Images

• Use high-resolution images for best quality
• Clear separation between subject and background helps
• Good lighting improves results

Model Selection

• General photos: Start with u2net
• People/faces: Use u2net_human_seg
• Professional work: Use isnet-general-use with resize_for_speed=false

Speed vs Quality

• Enable resize_for_speed for batch processing
• Disable resize_for_speed for final/professional output
• isnet-general-use gives best edges but takes longer

Deployment Ready

This model is optimized for production deployment: - Models pre-download during build for instant startup - Efficient memory usage - CPU-only for reliable scaling - Smart caching reduces processing time

API Response

Returns a PNG image with transparent background. The removed background is fully transparent, making it easy to composite onto any new background.

Credits & Licenses

This model uses the following open-source projects:

• rembg by Daniel Gatis - MIT License
• U-2-Net by Xuebin Qin et al. - Apache 2.0 License
• IS-Net by Xuebin Qin et al. - Apache 2.0 License

License

This implementation is provided as-is for educational and commercial use. The underlying AI models retain their original licenses as listed above.