Who bg0 is for#
Designers processing confidential client assets
Removes backgrounds from NDA-covered work, unreleased product mockups, or client portraits without sending files to a third-party service. The browser-based model means no data leaves the machine even in office environments with strict data policies.
Skip if:
The images are not sensitive and your workflow already approves use of cloud image tools.
Developers embedding background removal in web apps
The @bg0/browser package integrates local background removal in one function call. It manages WebGPU or WebAssembly detection, model caching, inference, and PNG output without any API key or dependency on bg0.dev uptime.
Skip if:
The app targets a server-side or non-browser runtime. @bg0/browser requires a browser environment with WebGPU or WebAssembly support.
Individuals removing backgrounds from personal photos
Processes ID photos, personal portraits, or sensitive images without any cloud upload. Opening the network panel during a run confirms that no pixel data leaves the browser, which is verifiable without trusting a privacy policy.
Skip if:
Privacy is not a concern and a free tier on an existing service handles the volume needed.
The problem it solves#
Background removal services process images on remote servers. For most workflows that is fine, but it creates a real constraint for teams under NDA, photographers handling client portraits, or developers building products where images must stay on-device.
The mainstream tools in this space (remove.bg, PhotoRoom, Clipping Magic) all require uploading to a server. You accept their terms of service, create an account, and depend on each service remaining available and priced reasonably. Offline processing is not an option; there is no way to run background removal without sending your pixels to a third party.
How it solves it#
WebGPU inference with WebAssembly fallback
BiRefNet-lite runs on the device's GPU through WebGPU for hardware-accelerated inference. Browsers without WebGPU fall back to WebAssembly automatically. The website records 1.8 seconds to process a 768x512 image on WebGPU. No server round-trip is involved in either path.
Zero network requests during processing
The browser's network panel shows no outbound requests containing image data. Pixels, filename, dimensions, and the processed result all stay in browser memory from start to finish. Close the tab and the image is gone.
Embeddable npm library
The @bg0/browser package exposes a removeBackground function that handles capability detection, model loading and caching, WebGPU or WebAssembly inference, and alpha compositing. Application code does not need to interact with ONNX internals directly.
Keyboard-first workflow
Every step from file selection to export has a shortcut. Open with ⌘O or ⌘V, hold Space to view the original, use arrow keys for fine comparison control, save with ⌘S, and copy to the clipboard with ⌘C. No mouse required.
No account, API key, or usage cap
There is no registration and no usage quota. The model downloads once and caches in the browser; subsequent processing runs without network access after that initial download. The Apache-2.0 license allows commercial use of both the site source and the @bg0/browser package.
Strengths and trade-offs#
Strengths
- Structural privacy guaranteeRunning inference in the browser eliminates the option to send pixels to a server. There is no backend endpoint for image data. Opening the network panel during any processing run shows no outbound requests carrying pixel data, filename, or metadata.
- Apache-2.0 with an embeddable libraryThe full source is Apache-2.0 licensed. The @bg0/browser package can be embedded in any web product without usage fees, API keys, or rate limits. Unlike remove.bg or PhotoRoom, which require an account and charge per image or by subscription, there is no per-use cost.
- Works offline after the first runThe BiRefNet-lite model caches locally after the initial download. Subsequent sessions process images without any internet connection. There is no API to go offline, no vendor pricing change to absorb, and no rate limit to hit during extended work.
Trade-offs
- -WebGPU coverage is not universalWebGPU is available in Chrome 113+ and Edge 113+ but remains experimental or unsupported in some browsers, including most Safari versions and older Firefox builds. Those fall back to WebAssembly, which is slower. On hardware without a capable GPU, processing times will exceed those seen on WebGPU.
- -Very early-stage projectbg0 was created on September 14, 2026, and had 57 stars at the time of listing. It has had no extended production exposure, and there is no track record for edge cases such as unusual image formats, very large files, or complex subjects that challenge the model.
- -No manual refinement in the web UIThe web app produces an automatic cutout and offers no brush tool or correction layer. If the BiRefNet-lite model makes an error on a complex edge (detailed hair, transparent objects, overlapping subjects), there is no in-app way to fix it. Clipping Magic is an option for work that needs manual correction.
bg0 vs alternatives#
bg0 vs remove.bg
Remove.bg is a cloud-based background removal service. It processes images on remote servers, requires an account and API key, and charges for usage. bg0 processes images locally in the browser with no API key, no account, and no usage limit.
| Feature | bg0 | remove.bg |
|---|---|---|
| License | Apache-2.0 | Proprietary |
| Processing | Local, in-browser | Cloud server |
| Cost | Free | Paid |
| API key required | No | Yes |
| Image stays on device | Yes | No |
Remove.bg is worth considering when cutout quality on complex subjects (detailed hair, fur, transparent objects) is the priority and cloud processing is acceptable. Remove.bg has years of production refinement; bg0 was released in September 2026 and is still early-stage.
bg0 vs PhotoRoom
PhotoRoom is a photo editing platform that includes background removal alongside replacement backgrounds, shadow effects, and batch editing tools. It processes images in the cloud and targets e-commerce and social media workflows. bg0 is a single-purpose background remover.
| Feature | bg0 | PhotoRoom |
|---|---|---|
| License | Apache-2.0 | Proprietary |
| Processing | Local, in-browser | Cloud |
| Scope | Background removal only | Removal + editing suite |
| Cost | Free | Paid subscription |
| Embeddable | @bg0/browser (npm) | API (paid) |
PhotoRoom is the better choice for e-commerce teams that need a full editing suite with batch operations. bg0 is the better choice when privacy matters or you need an embeddable library with no per-use cost.
bg0 vs Clipping Magic
Clipping Magic provides server-side background removal with a manual brush tool for correcting automatically generated cutouts. bg0 produces automatic cutouts locally with no manual correction option.
| Feature | bg0 | Clipping Magic |
|---|---|---|
| License | Apache-2.0 | Proprietary |
| Processing | Local, in-browser | Cloud server |
| Manual refinement | None | Brush tool |
| Cost | Free | Paid |
Clipping Magic is the better choice when automatic cutout quality is insufficient and you need to fix edges by hand. bg0 has no in-app correction path. For images where the automatic result is clean, bg0 is free, private, and requires no account.
Quick start#
Install the @bg0/browser package with npm to embed local background removal in any web application.
```bash
npm install @bg0/browser
```What it's built on#
- Languages
- TypeScript
- Frameworks
- React
FAQ#
Does bg0 upload images to a server?
No. bg0 runs the BiRefNet-lite ONNX model entirely in the browser. Opening the network panel during processing shows no outbound requests carrying pixel data, the filename, or the processed result. The model is the only thing that downloads, and it caches after the first use.
How does bg0 compare to remove.bg?
Remove.bg processes images server-side, requires an account and API key for programmatic access, and charges for usage. bg0 processes images in the browser with no API key, no account, and no usage limit. The trade-off is that bg0 is newer and may produce less consistent cutouts on complex subjects than remove.bg, which has years of production refinement.
Can I use @bg0/browser in my own application?
Yes. Install the package with npm and call removeBackground with a File or Blob. The package handles WebGPU or WebAssembly selection, model loading and caching, ONNX inference, and PNG output. The Apache-2.0 license permits commercial use.
Which browsers does bg0 support?
bg0 runs in any browser with WebGPU or WebAssembly support. WebGPU delivers faster inference and is available in Chrome 113+ and Edge 113+. Browsers without WebGPU, including most Safari versions and older Firefox builds, fall back to WebAssembly automatically.
Does bg0 require an account or API key?
No. The web app at bg0.dev works without registration, and the @bg0/browser npm package works without any API credentials. The BiRefNet-lite model downloads on first use and caches locally; there are no usage limits or subscription requirements.
Similar open-source tools#
God's Eye View
Real-time global intelligence on a 3D globe
hyperframes
Turn HTML and CSS into rendered MP4 videos, built for AI agents.
garden-skills
Production-ready agent skills for Claude Code, Cursor, and Codex
design.md
Generate UIs effortlessly with AI.
Fluent-Qt
Fluent UI component library for Qt Widgets in C++17 and Python
opendisplay
Turn any iPhone, iPad, or spare Mac into a free second Mac monitor