withoutBG

withoutBG is an open-source toolkit and hosted API for automatic image background removal. It provides lightweight local models for offline use (Focus and Snap), a Dockerized web UI, a Python SDK and a paid Pro API optimized for production inference.

Key Features

• Open-source Focus and Snap matting models (Apache-2.0) for local, privacy-preserving background removal
• withoutBG Pro API: higher-quality model variant running on AWS Inferentia for lower latency and improved edge detail
• Dockerized web app with drag-and-drop UI, CLI and Python SDK returning PIL Image objects for easy integration
• Pipeline combines depth estimation, saliency/segmentation signals and a downstream refiner to improve edge and hair/fur handling
• Small local model footprint (~320 MB total) with cached model downloads and batch-processing support for efficiency
• Privacy-first design: in‑memory processing, minimal logging, and explicit policy not to store or use customer images for training

Use Cases

• E-commerce and product photography: bulk-remove backgrounds for catalogs and marketplace listings
• Content creation and design: quick cutouts for thumbnails, compositing and UI assets
• Privacy-sensitive or offline workflows: run models locally for on-device or air-gapped processing

Limitations and Considerations

• Transparent or highly translucent materials (glass, water, sheer fabrics) can produce inconsistent results due to complex light interactions
• Cases with subjective foreground/background separation may produce results that differ from user expectations; very ambiguous scenes may need manual post-processing

withoutBG combines an easy-to-use developer workflow (CLI, Docker, Python SDK) with an option to scale via a hosted API. It is suitable for teams needing both private local processing and a production-ready hosted inference service.