Background Remover
Cut out the subject of a photo and get a transparent PNG, right in your browser, without uploading it.
Loading Background Remover… If nothing happens, please enable JavaScript.
Removing the background from a photo used to mean careful work with a pen tool, or uploading the picture to an online service and hoping it would not be kept. This tool does it on your own device instead. Drop in a photo of a product, a pet, a person or almost any object and a small neural network looks at the whole image, works out which pixels belong to the main subject, and turns everything else transparent. The result is a PNG with a real alpha channel, ready for a shop listing, a slide, a sticker, a profile picture or a design mock-up, and you can check it against the original with the before and after slider.
Frequently asked questions
Are my photos uploaded anywhere?
Which quality mode should I pick?
Why does the first run take a while?
How good are the edges around hair and fur?
Can I replace the background instead of making it transparent?
Which format should I download?
Is the output the same resolution as my photo?
Why is it slow on my computer or phone?
Can I use the results commercially?
About Background Remover
There are two quality modes. Best uses BiRefNet, a model built for separating a foreground from its background whatever the subject is, and handles objects, products, animals and people; the first time you use it the browser downloads about 94 MB from this site and keeps it in its cache, so later visits start straight away. Fast uses MODNet, a much smaller model of about 7 MB made only for portraits: it is quick and good at hair around a head and shoulders, but it does not understand objects at all, so use it only for photos of people. With a recent graphics card and a browser that supports WebGPU a photo takes about a second; on the CPU it takes several seconds, depending on the device.
Once the subject is cut out you choose what goes behind it: keep it transparent, fill it with a solid colour (handy for white backgrounds on marketplaces or a coloured ID-style backdrop), or use a blurred copy of the original for a soft portrait look. Two refine sliders let you harden a soft edge or feather a sharp one, and the result can be saved as PNG, WebP or JPEG at the original resolution. You can add many photos at once, flip through them one at a time, and download everything as a single ZIP file. Nothing is ever sent to a server: the models and the processing run entirely inside this browser tab.
From blue screens to neural mattes
Separating a subject from its background is one of the oldest problems in film. Early visual effects used travelling mattes, where actors were filmed against an evenly lit blue backdrop so a chemical process could isolate them; Petro Vlahos refined the colour-difference method in the late 1950s and won several Academy Awards for it. Green later replaced blue in digital work because camera sensors capture green with the least noise.
All of those techniques needed a controlled backdrop. Cutting a subject out of an ordinary photo, with a cluttered room or a street behind it, is much harder, because the computer must first understand what the subject is. Classic tools such as GrabCut (2004) asked the user to draw a rough box and then separated colours statistically, which worked on simple scenes and struggled elsewhere.
Modern models learn the idea of a main subject from large sets of hand-labelled images. MODNet, published in 2020, focused on portraits and was small enough for real-time video. BiRefNet, from 2024, combines a global view of the scene with a high-resolution refinement step and handles almost any kind of object. The lite version used here fits in about 94 MB, small enough to run inside a browser tab.