Removing a background is the step between a photograph and a product listing, a slide, a logo or a profile picture. This tool does it on your own device: the image is read into a canvas, separated, and handed straight back. Nothing is uploaded, which is why there is no queue, no file size cap imposed by a server, and no copy of your photo sitting in someone else's storage afterwards.
How to use the Background Remover
- Drop an image onto the upload area, or click to browse
- Check the preview — the checkerboard shows what has become transparent
- Adjust Tolerance if too much or too little was removed
- If the wrong colour was detected, click the preview to sample the real background
- Choose a transparent PNG or a solid colour, then download
What it is good at, and what it is not
This works by finding the backdrop and spreading inward from the edges of the frame, so it is strongest exactly where most people need it: a subject against a reasonably plain background. Product shots on white, logos, screenshots, a portrait against a wall or studio paper, a flat-lay on a desk — these come out clean.
It is not a trained segmentation model, and it is worth being plain about the difference. It cannot lift a person out of a crowded street scene, and it will not resolve every strand of flyaway hair against a busy background. Doing that well needs a neural network and a weights file of several megabytes, which would mean downloading a model at runtime — and that would break the thing that makes this tool worth using, which is that your image never leaves the page. If an image is a poor fit, the tool says so above the preview rather than quietly handing back a bad cut-out.
Why a gradient backdrop does not break it
The naive approach is to pick one background colour and delete every pixel close to it. That falls over on the most ordinary thing in real photography: a backdrop that shades from light to dark across the frame, or vignettes toward the corners. Widen the threshold far enough to catch the dark end and it starts eating the subject too.
Instead this spreads outward from the border and compares each pixel to the neighbour it spread from, so it can follow a gradient as far as the gradient goes, while a second check against the range of colours actually found around the border keeps it from wandering into the subject. A practical consequence worth knowing: a region the same colour as the background is kept if it is not connected to the edge of the frame. The hole in the middle of a ring stays filled unless it opens onto the outside.
Getting a better cut-out
Tolerance is the control that matters. Too low and patches of background survive; too high and parts of the subject vanish. The default sits in the middle with room to move either way — if the subject starts disappearing, bring it down.
If the backdrop is uneven or the automatic detection picks the wrong colour, click the preview to sample the colour you want treated as background. Edge softness feathers the cut so it does not look stamped out, and Clean edge colour removes the backdrop's tint from semi-transparent edge pixels — that is what stops a white halo appearing when you place the cut-out on a dark background.
Two things help more than any setting: photograph the subject against a plain wall or sheet of paper, and leave a margin so the subject does not run off the edge of the frame.
Frequently asked questions
Is my image uploaded anywhere?
No. The separation runs in your browser on your own device. There is no server endpoint here that could receive an image, which you can confirm by watching the network tab in your browser's developer tools while you use it. It also means the tool keeps working with no connection.
Why does it work on some photos and not others?
It finds the background by spreading inward from the edges of the frame, so it needs a backdrop that is reasonably consistent. A subject on a plain wall, a product on white, a logo or a screenshot will come out clean. A subject standing in a detailed scene has no single backdrop to find, and the tool will tell you so instead of guessing.
Does it handle hair and soft edges?
It feathers the edge and removes the background's colour from partly transparent pixels, which handles ordinary soft edges and fine detail against a plain backdrop well. It will not resolve every individual strand against a busy background — that needs a trained model, and running one would mean downloading megabytes of weights, which this tool deliberately does not do.
Why is part of my subject missing?
The tolerance is too high for this image, or the subject contains a colour close to the background. Lower the tolerance first. If the subject runs off the edge of the frame it can also be mistaken for background — a warning appears above the preview when that looks likely.
Why is there a faint outline of the old background?
That is colour contamination on the boundary pixels, where the original image blended subject and backdrop together. Leave "Clean edge colour" switched on and it is reversed mathematically, which is what keeps a white halo from showing up when you drop the cut-out onto a dark background.
What size images can it handle?
Up to 40 megapixels, which is larger than any current phone or consumer camera produces. A 12 megapixel photo takes about a second. There is no limit on how many images you process, because there is no server deciding your quota.
Can I put a different background behind the subject?
Yes — choose white, black or a custom colour under Output and the cut-out is composited onto it, giving you a fully opaque image rather than a transparent one. For a photographic background, download the transparent PNG and layer it in an editor.