BG0 vs unbagrnd
BG0
bg0.devunbagrnd
zidniryi.spaceSide by side
| Figure | BG0 | unbagrnd |
|---|---|---|
| Price from | Free | Free |
| Free plan | Yes | Yes |
| Free trial | — | — |
| Open source | Yes | — |
| Web | Yes | — |
| Windows | — | Yes |
| macOS | — | Yes |
| Linux | — | Yes |
| iOS | Yes | — |
| Android | Yes | Yes |
| API | Yes | Yes |
| Self-hosted | Yes | Yes |
| Plans published | 1 | 1 |
| Spec score · Image Background Removal Software | 9.0 (#5) | 7.5 (#21) |
| Spec score · AI Background Removers | 8.3 (#35) | 7.5 (#73) |
| Free plan | Yes | Yes |
| License | The source code is released under Apache License 2.0, while model weights have separate notices and licenses. | The product is free and open source under the MIT license, with no API keys, paid tiers, or usage limits. |
| Local processing | Image bytes, filenames, thumbnails and masks remain on the device instead of being sent to a processing server. | The desktop app runs background removal locally with ONNX Runtime, and says images, filenames, and metadata are never sent anywhere. |
Plans
In their own words
BG0
BG0 is a free web tool that removes image backgrounds in the browser and downloads transparent PNG files. Image bytes remain in the browser rather than being uploaded for processing. The remover accepts PNG, JPG, WebP, HEIC, and HEIF images up to 40 MB. It runs BiRefNet and can fall back to BiRefNet-lite when the selected model cannot run. First use requires an internet connection to download and prepare the model, after which the browser may cache it. There is no account, credit system, subscription, or stated usage limit. Developers can use the @bg0/browser library to add on-device removal to web apps without a server or API key; it runs through WebGPU or WebAssembly. The documented iOS path uses a single-threaded WebAssembly worker and BiRefNet-lite, with output limited to 1280 pixels on the longest edge. No manual touch-up editor is included. BG0 uses PostHog for anonymous analytics.
unbagrnd
unbagrnd removes photo backgrounds with on-device AI, processing images locally in its desktop app with ONNX Runtime. It can handle a single image or a folder, with live before-and-after previews as processing completes. Users can pick from several models to balance speed and accuracy; the default is 43 MB. Brush tools erase or restore subject areas, with undo and redo, and users can add a solid-color background or an on-device drop shadow. The app predicts a foreground mask on the CPU, combines it with the original pixels, and saves the result as PNG. The maker says the only network request is a one-time model download, with no analytics, update checks, or telemetry. unbagrnd is free and open source under the MIT license, with no API keys, paid tiers, or usage limits. A self-hosted, Docker-capable REST API supports single-image and batch processing. Its documented defaults allow 20 MB per image, batches of up to 10, and 60 requests per minute per API key. It supports Android, Linux, macOS, Windows, and self-hosted use.