Remove Background from Images

Remove the background from photos automatically. Get a transparent PNG or a white or coloured background, in batches, right in your browser.

Your files are processed in your browser and never uploaded.

New background

85

Higher means better quality and larger files.

Works best on photos with one clear subject — a person, a product, a pet, a car. Open each result to check the edges.

The AI model (about 5 MB) downloads once, then stays in your browser's cache.

Drop images here

Choose files

HEIC, JPG, PNG, WebP, AVIF, GIF, BMP · up to 50 megapixels each · you can also paste

How to use

  1. Choose the new background (transparent, white or any colour) and the output format.
  2. Drop your photos or click "Choose files". The subject in each photo is found and cut out automatically.
  3. Open a result to check the edges, then download the images one by one or all together as a ZIP.

Features

Fully automatic
No brushing, no clicking around the outline. An AI model finds the main subject and the edge is refined at the photo's full resolution.
Clean edges, no halo
Along soft edges such as fur and hair, the old background colour is removed from the pixels themselves, so a cat shot on snow doesn't keep a white rim on a dark background.
Transparent, white or any colour
Save a transparent WebP or PNG for design work, or put the subject on white or a brand colour and save a smaller JPG.
Private and unlimited
Everything runs on your device, so there is no upload, no account, no credit system and no reduced-resolution "free preview". Only the model itself is downloaded, once.

How automatic background removal works

The tool uses U²-Netp, an open-source salient object detection network published by researchers at the University of Alberta. It is trained to answer one question: which part of the picture would a person look at first? For each photo, a 320 × 320 pixel copy is passed through the network, which returns a map of how likely each point is to belong to the main subject. That map is the starting point, but on its own it is far too coarse for a 12-megapixel photo: scaled straight up, every edge would be a blurry band dozens of pixels wide.

Refining the edge at full resolution

To get a sharp outline, the map is refined with a guided filter. In every small neighbourhood of the image, the filter finds the simplest rule that turns the photo’s colours into the subject map, then applies that rule to the full-resolution pixels. Where the subject and background differ in colour, the edge snaps to the real boundary; where the model was unsure, the edge stays soft rather than jagged. This step is computed on a copy about 1,000 pixels wide and applied to the full image, so it takes well under a second.

Removing colour fringes

A pixel on the edge of a cat’s fur is partly cat and partly whatever was behind it. If it keeps its original colour, the subject carries a thin outline of the old background onto the new one: a white halo from a studio backdrop, a green tint from grass. The tool estimates the subject’s own colour and the background’s colour around each edge pixel and removes the background’s share, which is what makes results on a coloured background look natural.

Tips for better cut-outs

Shoot against a plain background that contrasts with the subject, keep the subject in focus, and leave a little space around it. Crop away clutter before processing, for example with the crop tool, so the subject fills more of the frame. If you need a specific size for a shop listing or social media, run the transparent PNG through Resize Image afterwards; resizing keeps the transparency.

Frequently asked questions

What kind of photos work best?

Photos with one clear main subject that stands out from what is behind it — a person, a product, a pet, a car, a plant in a pot. Product shots on a plain table or wall, portraits and pet photos usually come out clean. The result is weaker when the subject has a colour very close to the background, when several things compete for attention, or in wide scenes such as streets and rooms where there is no single subject.

Why are some edges soft or slightly wrong?

The model looks at a reduced copy of the photo to decide what the subject is, and the outline is then fitted to the colour edges at full resolution. Wisps of hair, whiskers, glass and lace are only partly captured, and a shadow under an object may be kept or removed depending on how dark it is. Open each result to check it; for important images you can touch up the edge in an editor afterwards.

Which format should I choose?

For a transparent background the default is WebP. A 12-megapixel cutout is usually around 1 MB, and current browsers and most design tools open it; the quality slider trades size against detail. Choose PNG when an older app needs it. PNG is lossless, so the same cutout is often 8–14 MB, several times the original photo. With a white or coloured background JPG is the usual choice because it is the smallest. JPG cannot store transparency.

Can I use it for passport or visa photos?

Usually not. Most passport and visa authorities do not accept photos whose background has been digitally replaced and ask for a plain, light background at the time the photo is taken. Take the photo against a white or light grey wall and use our [passport photo tool](/en/passport-photo/), which only crops and resizes.

Are my photos uploaded?

No. The page downloads the AI model and the image code to your browser, and every photo is processed there. Once the first photo has finished, you can even go offline and the tool keeps working.

Is there a limit on how many images I can process?

No daily limit and no watermark. Photos are processed one or two at a time to keep your device responsive, and each photo can be up to 50 megapixels. A 12-megapixel photo takes a few seconds on a recent laptop and longer on a phone.

Related tools