Erasemate
AI-Powered Background Removal with Advanced Alpha Matting
Erasemate is a professional-grade AI-powered background removal system featuring both a web application and REST API for developers. Built with cutting-edge deep learning models, it automatically removes image backgrounds while preserving fine details through advanced alpha matting—capturing individual hair strands, fur textures, and semi-transparent edges that standard segmentation tools discard. The system employs intelligent model selection, routing portrait images through a dedicated human segmentation pipeline and general subjects through optimized architectures. Key features include automatic image downscaling and upscaling for handling large inputs, batch processing for up to 10 images per request, optional background color compositing, and JWT-based authentication with row-level security. The backend leverages U2Net, U2Net Human Seg, and ISNet deep learning models for state-of-the-art inference, with per-user daily usage tracking and edge refinement via median filtering and Gaussian blending. The frontend provides a clean, responsive interface built with React and TypeScript, while the backend is powered by FastAPI with Supabase for authentication and PostgreSQL for usage tracking. The architecture supports streaming responses for single images and JSON responses for batch operations. Deployed on Vercel (frontend) and Hugging Face Spaces (backend with 16GB RAM), it offers OpenAPI documentation and handles up to 25MB file uploads with configurable rate limiting for free-tier users.