Remove image background

A practical guide to removing image backgrounds

Removing a background sounds like one action, but it is really two: deciding which pixels belong to the subject, then deleting everything else without leaving a halo or chewing up the edges. PixTools handles the first part with a segmentation model that runs in your browser, so the whole job happens on your own device. This guide explains what the tool is doing, when an automatic cutout will look clean, and how to get a usable result on the first try. For a fuller walkthrough, see how to remove an image background.

How the cutout is made on your device

When you choose a file, your browser downloads the segmentation model the first time and then runs it locally. The model estimates a mask — a map of how likely each pixel is to be part of the main subject — and that mask becomes the transparent alpha channel of the exported PNG. Because the model and the image both stay in the browser, your photo is not sent to a PixTools server for the cutout. The trade-off is that the first run is slower while the model loads, and very large images take more memory and time than small ones.

Why some images cut out cleanly and others do not

Automatic removal is strongest when the subject is clearly different from what is behind it: a person against a plain wall, a product on a seamless backdrop, a pet on grass. It struggles where the boundary is genuinely ambiguous — fine flyaway hair and fur, motion blur, glass and other see-through objects, soft shadows that read as part of the subject, and busy backgrounds that share the subject's colors. In those cases the model has to guess, and you may see a rough edge, a missing strand, or a piece of background left behind. None of this is a bug; it is the limit of deciding edges from a single flat image.

Transparent PNG or a replacement background?

Export a transparent PNG when the cutout will be placed on top of something else: a design layout, a slide, a sticker, a layered graphic, or a thumbnail where you want to add your own backdrop later. Choose a solid white, black, or custom color when you need a finished, self-contained image — for example, marketplace and catalog listings that require a clean white background, or profile pictures that look tidier on one flat color. If you are unsure, keep the transparent version; you can always drop it onto a color later, but you cannot recover transparency once it has been flattened.

Getting a cleaner cutout from the start

The single biggest improvement is contrast between the subject and the background before you ever upload the file. Shoot or pick an image where the subject is sharp, well lit, and clearly separated from what is behind it, and avoid backgrounds that match the subject's color or tone. A little extra resolution helps the model find edges, but enormous files mostly slow things down without improving the mask, so a sensibly sized image is usually the sweet spot. If an edge looks wrong, re-cropping to give the subject a bit of breathing room and re-running often helps more than fighting a difficult original.

Finishing the exported image

The cutout is rarely the last step. If the PNG is larger than you need for the web, run it through the Image Resizer to bring down the dimensions and file weight while keeping the transparency. If a destination expects a specific file type, the Image Converter can turn the PNG into WebP or other outputs. For a flattened JPEG copy, the Image Compressor can reduce file weight without changing dimensions. And if the same photo also shows people whose faces should not be public, run it through Face Blur before you publish or share the final image.