---
title: BiRefNet Background Remover
emoji: ✂️
colorFrom: indigo
colorTo: purple
sdk: gradio
sdk_version: 5.45.0
app_file: app.py
pinned: false
license: mit
short_description: Remove image backgrounds with BiRefNet.
---

# BiRefNet Background Remover

A small Gradio app that creates transparent PNG cut-outs with
[BiRefNet-General](https://github.com/ZhengPeng7/BiRefNet), through
[rembg](https://github.com/danielgatis/rembg). It replaces the previous
RMBG-2.0/Transformers implementation.

## Run locally

Requires Python 3.10 or newer.

```bash
python -m venv .venv
# Windows PowerShell
.venv\Scripts\Activate.ps1
pip install -r requirements.txt
python app.py
```

Open the local URL printed by Gradio. The first removal downloads the
`birefnet-general` ONNX weights and caches them; later requests reuse the
same in-process session. To choose a persistent cache location, set
`REMBG_HOME` before starting the app.

## Deploy to Hugging Face Spaces

1. Create a new **Gradio** Space, preferably with Python 3.10+.
2. Push these files to the root of the Space: `app.py`,
   `remove_background.py`, `requirements.txt`, `README.md`, and `LICENSE`.
3. Wait for the build to finish. No secrets or manual model download are
   required. The model downloads automatically on the first inference.
4. For quicker or more reliable cold starts, use a persistent-storage upgrade
   and set `REMBG_HOME=/data/rembg` in the Space variables. This is optional;
   the default cache also works but is recreated after a free Space restarts.

The app queues one request at a time to keep memory usage safe. It supports
Hugging Face ZeroGPU and requests a GPU only while removing a background; a
CPU Space also works, albeit more slowly.

## Licensing and attribution

This repository's code is MIT licensed. BiRefNet's upstream repository is
MIT licensed, and `rembg` is MIT licensed. The selected `birefnet-general`
model is the BiRefNet general-use ONNX model distributed by `rembg`; unlike
the prior BRIA RMBG-2.0 model, it is intended here to support commercial use.
Keep this attribution and independently review upstream model and dataset
terms for your specific deployment.

Links: [BiRefNet](https://github.com/ZhengPeng7/BiRefNet) ·
[rembg model list](https://github.com/danielgatis/rembg#available-models)