Technical Deep Dive

Learn how cutting-edge AI and WebAssembly technology power our background removal tool

The Technology Behind the Magic

Enterprise-grade AI running entirely in your browser through cutting-edge WebAssembly technology

🧠

ISNet Neural Network

Intermediate Supervision Network is a state-of-the-art deep learning architecture designed specifically for salient object detection and segmentation.

•Multi-scale feature extraction: Analyzes images at multiple resolutions for precise edge detection
•Deep supervision: Intermediate layers guide learning for better accuracy
•Encoder-decoder architecture: Captures both high-level semantics and fine details
•Trained on DIS5K dataset: 5,470 high-quality images with pixel-perfect masks

⚡

ONNX Runtime WASM

Open Neural Network Exchange Runtime compiled to WebAssembly enables near-native performance for AI inference directly in the browser.

•Hardware acceleration: Leverages SIMD instructions for parallel processing
•Optimized operators: Specialized implementations for convolutions and pooling
•Memory efficient: Streams processing to handle large images
•Cross-platform: Runs identically across all modern browsers

🔄 AI Processing Pipeline

📥

1. Image Input

Decode & normalize pixel data

🔍

2. Feature Extract

CNN layers detect patterns

🎯

3. Segmentation

Generate alpha mask

✂️

4. Mask Apply

Separate foreground

💾

5. Encode Output

PNG/WebP with alpha

📊 Model Variants Comparison

Model	Precision	Size	Speed	Accuracy	Best For
ISNet Full	FP32 (32-bit)	~20 MB	15-30s	⭐⭐⭐⭐⭐	Professional work, complex subjects
ISNet FP16	FP16 (16-bit)	~10 MB	8-15s	⭐⭐⭐⭐	Balanced quality & performance
ISNet Quint8	INT8 (8-bit)	~5 MB	3-8s	⭐⭐⭐⭐	Bulk processing, quick results

Key Features & Benefits

🔒

100% Private & Secure

All AI processing happens locally in your browser. Your images never leave your device. No server uploads, no data collection, complete privacy guaranteed.

⚡

WebAssembly Speed

ONNX Runtime compiled to WASM provides near-native performance. Process images 10x faster than JavaScript-only solutions with SIMD acceleration.

🎯

State-of-the-Art AI

ISNet neural network trained on thousands of images delivers professional-grade results. Handles complex hair, transparent objects, and fine details with precision.

🌐

Works Offline

Once the model is loaded, it's cached in your browser. Process images without an internet connection. Perfect for travel or unreliable networks.

💰

Completely Free

No subscriptions, no credits, no limits. Process unlimited images for free. Open-source technology powered by @imgly/background-removal library.

📦

Batch Processing

Process multiple images in sequence with detailed progress tracking. Automatically download all processed images with one click.

Technical Specifications

🌐 Browser Requirements

✓WebAssembly Support: Chrome 57+, Firefox 52+, Safari 11+, Edge 16+
✓SIMD Instructions: For optimal performance (Chrome 91+, Firefox 89+)
✓Minimum RAM: 4GB recommended for smooth processing
✓Storage: ~20-30MB for model caching (IndexedDB)

📄 Supported Formats

Input Formats:

JPEG, PNG, WebP, BMP, TIFF

Output Formats:

PNG (24-bit + 8-bit alpha), WebP (with alpha), JPEG (no transparency)

Max Image Size:

4096×4096 pixels (auto-resize for larger images)

Color Space:

RGB, sRGB color space

📈 Performance Metrics

Model Load Time (first visit):10-30s

Model Load Time (cached):< 1s

Processing (1024×768, Quint8):3-5s

Processing (1024×768, Full):15-20s

Memory Usage:200-500MB

🛠️ Technology Stack

▸Framework: Vue 3 Composition API with Nuxt 3
▸AI Library: @imgly/background-removal v1.4.5+
▸Runtime: ONNX Runtime Web (WASM + SIMD)
▸Styling: Tailwind CSS 3.x
▸Model: ISNet (DIS5K trained)

Ready to Try It Out?

Experience the power of AI-powered background removal in your browser

Start Removing Backgrounds