Background Remover

How to remove the background from a photo

1. Drop a photo on the drop zone, click it to choose a file, or paste an image with Ctrl+V (⌘V on a Mac).
2. The tool starts right away. The first time, it downloads its AI model and engine; after that, a photo takes a few seconds. You can also click Remove background to run it again.
3. Check the result with View: Before / after shows a slider (drag Before ← → After to compare), Result shows the cut-out on a checkerboard, and Mask shows what is kept in white and removed in black.
4. If the edges need work, use Refine the mask (see below).
5. Under New background, keep Transparent, or pick Color, Image (drop a second picture) or Blurred original.
6. Click Download PNG. With Transparent, the PNG keeps an alpha channel, ready for a slide, a product page or a design tool.

How it works

The tool uses U²-Net (the compact “U²-Netp” version), a neural network published by Xuebin Qin and colleagues in 2020 for salient object detection: it finds the main subject of a picture, whether a person, an animal, a product or an object, and gives every pixel a probability of belonging to it. The network runs on your own device with ONNX Runtime compiled to WebAssembly. Your photo is never sent anywhere: the model file comes from this site, and all computing happens in your browser tab.

The network works on a 320 × 320 version of the photo. The probability map is then scaled back to the full size of your image (up to 4096 px on the longest side; larger photos are scaled down first) and used as the transparency of each pixel.

Refining the mask

• Threshold decides how sure the model must be before a pixel is kept. Raise it to remove leftover bits of background; lower it if parts of the subject disappear.
• Edge softness sets how gradual the transition between kept and removed pixels is. At 0 % you get a hard cut-out (useful for logos and flat objects); higher values give natural edges for hair and fur.
• Feather blurs the final edge by a few pixels, which helps a cut-out blend into a new background.

The Mask view is the easiest way to judge these settings.

Tips and limits

• Photos with one clear subject in front of a distinct background work best: portraits, pets, products on a table, cars.
• Busy scenes with several subjects, or a subject that has the same colors as the background, can be partly missed. Try another Threshold, or crop the photo around the subject first.
• Very fine details such as flyaway hair or a glass’s transparency are approximated: the compact model favors speed and a small download.
• The download is always a PNG. PNG is lossless and supports transparency; if you need a smaller JPG with a solid background, convert the PNG afterwards.