withoutbg-python
Python SDK for local and cloud background removal using open weights or cloud API
Switch between free local open weights and a sharp cloud API with a single line of code
withoutbg-python is a Python SDK for removing image backgrounds using either local open-weights ONNX models or a cloud API. It suits developers building automated scripts, notebooks, backends, or batch jobs that need flexible execution options.
What it is
The tool provides a unified API to run background removal locally using a router between DINOv3-fused matting and BiRefNet segmentation, or route requests to a cloud API. It returns a PIL Image in RGBA format and supports batch processing with custom progress callbacks.
Who it's for
It suits Python developers and technical photographers building custom image processing pipelines or batch workflows. It does not suit non-technical users looking for a standalone desktop application without writing code.
Key features
Unified API Design
Switch between local open weights and the cloud API with a single line of code.
Batch Processing Support
Process multiple images efficiently while keeping the model loaded in memory.
Progress Callbacks
Track processing progress with custom callback functions for long-running jobs.
Multi-Branch Router
Intelligently routes images between matting and segmentation models based on content.
Up and running in 3 steps
Our honest verdict
What's great
- Identical API for both local offline execution and cloud processing
- Supports native PIL Image integration for easy compositing in Python workflows
- Includes a command-line interface for quick terminal-based background removal
- First local run automatically downloads necessary ONNX weights from Hugging Face
Where it falls short
- First local run requires downloading roughly 1.5 gigabytes of model weights
- Local open weights require maintaining sidecar metadata files if using custom paths
- Cloud API mode requires a paid API key for processing images
Pricing & license
The Python SDK and local open weights are free to use under the Apache 2.0 license, while the optional cloud API requires paying per image processed.
Open remote AI-training roles
768 open remote AI-training roles, updated November 14, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
Alternatives
These reviewed AI tools do the same job instantly — from our scored directory.
Use it, or have us set it up
FAQ
Can I run this completely offline?
Yes, once the initial 1.5 GB model weights are downloaded from Hugging Face, local open-weights mode runs entirely offline.
What image formats are supported for output?
The SDK returns a PIL Image in RGBA format, which can be saved as PNG or WebP to preserve transparency.
How do I handle batch processing efficiently?
Initialize the model once outside your loop and use remove_background_batch() to avoid reloading weights for every image.
