Background Remover

What does the Background Remover do?

It cuts out the main subject of a photo, such as a person, a product, an animal or a car, and makes everything else transparent. The result is a PNG with an alpha channel that you can put on any background, use for a product listing, a profile picture, a slide or a thumbnail. You can also replace the background with a solid color, for example white for an online shop.

The segmentation model, BiRefNet_lite, runs inside your browser with WebAssembly. Your photo stays on your device.

How to use

1. Drop an image onto the drop area, click it to choose a file, or paste an image with Ctrl+V (Cmd+V on a Mac).
2. Wait while the AI model is downloaded (about 113 MB the first time you use it on a visit) and the background is removed. This usually takes 5 to 20 seconds.
3. Drag across the preview to compare the Original and the Result. The transparent area is shown as a checkerboard.
4. Choose Transparent or Color for the background. With Color, pick any color or one of the presets.
5. Click Download PNG. The file is saved as name-no-bg.png.

To process another image, drop it onto the preview or click Choose another image.

Tips for good results

• Photos with one clear subject work best: a person, a product, a pet, a car.
• Good contrast between the subject and the background helps, especially around hair and fur.
• Very large images are reduced to about 16 megapixels (4096 × 4096) before processing.
• The model looks at the image at 512 × 512, and the mask is scaled up to the full size. Edges are soft and natural, but very thin details such as single hairs or wires may be lost.

Limitations

• The model decides what the subject is. If a photo has several objects or a busy scene, it may keep or remove parts you did not expect. There is no manual brush to fix the mask.
• Transparent and reflective objects (glass, water) are hard for any automatic tool.
• One image at a time.
• The model needs a fair amount of memory (about 2 GB while it runs). On phones with little memory the tab may reload.

FAQ

Are my images uploaded?

No. The image is decoded, processed and encoded in your browser. Only the model and the ONNX Runtime engine are downloaded from public CDNs (Hugging Face and jsDelivr); nothing is sent back. The image and the result are saved in this browser's storage with the tool (up to 20 MB in total), so they're still there after a reload. If you share the tool with a link, they're included in the link. To not keep them, turn off Keep files in the tool's settings.

Is it free? Is there a watermark or size limit?

It is free, there is no watermark, and the PNG has the same size as your image (up to about 16 megapixels).

Which AI model is used, and under what license?

BiRefNet_lite by Peng Zheng et al., released under the MIT License. The tool uses a 512 × 512 ONNX export of it (also MIT, about 98 MB in half precision), run with ONNX Runtime Web (MIT). Every downloaded file is pinned to a fixed version and checked against a SHA-256 hash before it is used.

Why does it take a while the first time?

The model (about 98 MB) and the ONNX Runtime engine (about 14 MB) have to be downloaded. They stay in memory while the page is open, so the next images are processed much faster. Your browser may also keep them in its cache for later visits.

Can I get a white background instead of a transparent one?

Yes. Choose Color and pick white (or any other color). The downloaded PNG then has the color filled in behind the subject.

Does it work offline?

No. The model has to be downloaded the first time on each visit (unless your browser has cached it).