Remove Background from Image — Free, Private, Full Resolution

Full-resolution background removal with on-device CPU inference, exported as a transparent PNG.

Separating a subject from whatever happened to be behind it is a segmentation problem, and this tool gives you three different ways to solve it inside your own browser. Pick the method that suits the photo, tidy the mask edge with threshold, shrink and feather controls, touch up by hand where needed, and export a cutout with real transparency.

How to remove a background from a photo

  1. Drop a JPG, PNG or WebP photo onto the upload area, or click it to pick one from your device.
  2. Choose Auto for most subjects, People & portraits for a person, or Flat background for a plain backdrop; the first cutout appears over a checkerboard preview.
  3. Refine the mask with Hard edge, Shrink / grow (up to 10 px either way) and Feather (up to 20 px); in Flat mode, adjust the tolerance or pick the backdrop colour yourself.
  4. Drag across the preview with the Erase or Restore brush, sized from 4 to 60 px, to fix hair, fingers or a leftover shadow.
  5. Click Download full-resolution PNG to save the cutout with a transparent background.

Three methods for three kinds of photo

Auto runs U²-Netp, a nested encoder-decoder trained for salient-object detection. At roughly four and a half megabytes it downloads once and is cached, and it handles the widest range of subjects: products, pets, furniture and people alike.

People & portraits runs MediaPipe's selfie segmenter, a quarter-megabyte model trained only on humans. On a head-and-shoulders shot it follows hair and collars more closely than a general saliency model, but it ignores anything that is not a person.

Flat background uses no model at all. It finds the dominant colour around the photo's edge (or the colour you pick), then floods every connected pixel within your tolerance. For catalogue shots on a seamless white sweep or app screenshots it is instant and pixel-exact.

Everything runs on your processor

All three methods run in a background worker on the CPU through WebAssembly, so the page stays responsive while the mask is computed and no graphics driver quirk can change the result. A large photo is analysed at a working size of up to 1600 pixels and the finished mask is scaled back onto the full-resolution original.

Switching method re-runs the cutout on the same photo, so when the first attempt misses a limb or keeps a shadow, trying another approach costs a few seconds rather than a new upload.

Shaping the edge

Hard edge turns the soft mask into a crisp one at the level you choose, which suits logos and product shots. Shrink or grow moves the boundary inward to drop a coloured fringe or outward to rescue fine detail, and feather blurs the transition so the cutout sits naturally on a new background.

After that the brush lets you paint areas back in or carve them away at whatever radius suits the detail. Exports carry a genuine alpha channel, so the cutout drops onto any background later without a white rectangle around it.

Free background removal at full resolution

Photos that suit each background removal method

Questions people ask about Background Remover

Is my photograph uploaded anywhere?
No. The image is decoded in the page, handed to a background worker, and never leaves the device. Even the model weights come from this site's own address rather than a third-party host, so nothing about your picture reaches another server.
Why did it cut off part of my subject?
Each method has blind spots: saliency may skip a limb that blends into the background, the selfie model ignores objects a person is holding, and colour flood removes anything that matches the backdrop. Try another method, grow the edge a few pixels, or paint the missing region back with the brush.
Why does the cutout always download as a PNG?
Transparency is the reason to make a cutout, and PNG stores a full alpha channel without compression artefacts, so it is the only export offered. If you need a JPG on a white backdrop, place the PNG over a filled layer in any image editor and save it from there.
Does it work on more than one subject in a frame?
Often, since the model marks salient regions rather than counting them, so two people standing together usually come through as one combined foreground. Separating them from each other is manual work with the brush afterwards.
Which method should I start with?
Use Flat background when the backdrop is one plain colour, People & portraits for photos of a person, and Auto for everything else. The method can be changed at any time and the photo is processed again without reloading it.
Can I remove a background at full or HD resolution for free?
Yes. The mask is worked out on a copy of up to 1600 pixels on the long side, then scaled back up and applied to your original, so the PNG you download has exactly the source's width and height with no paid upgrade involved.
Can I remove the background without losing image quality?
The subject's pixels are copied from your original photo and saved as a lossless PNG, so colour and sharpness are unchanged and only the alpha channel is new. On very large photos the mask edge is upscaled from the working copy, so use feather, shrink or the brush where a boundary looks soft.
Does the background remover work on a phone?
It needs a browser with Web Workers and WebAssembly, and the touch-up brush responds to a finger as well as a mouse. Expect slower processing than on a desktop processor, and note that the 4.6 MB Auto model downloads once over your connection the first time you use it.

Related tools on BrakeTools

Content reviewed 2026-09-27.