Background Remover

How the background is detected

The tool does not know what your subject is, so it works the way a magic wand does. First it averages the colour in the four corners of the image to estimate the backdrop. Then it starts at every pixel along the border and flood-fills inward, marking each connected pixel whose colour is within the tolerance of that estimate. Whatever the fill reaches becomes background; whatever it does not reach — normally the subject — is kept. The marked pixels have their alpha set to zero, so the output is a genuine transparent PNG rather than an image with the background painted white.

Tolerance and feather are the two controls that matter

Tolerance decides how far a pixel's colour may sit from the estimated background and still be erased. A low tolerance protects the subject but can leave specks of an uneven backdrop; a high tolerance clears a shaded or gradient background but risks eating into a subject that shares its colour. Feather blurs the alpha along the cut edge by a small radius. Without it the boundary is a hard, aliased line that looks pasted on and can carry a thin rim of the old background; with a little feather the subject blends onto whatever you place behind it. Because both are sliders, the workflow is usually run, look, nudge, run again.

Where a colour matte wins and where it does not

This approach is fast, private and very good at the common case: a product on white, a person against a smooth studio wall, a logo on a solid colour. It is the wrong tool for hair, fur, glass, fine mesh, heavy shadow, or a busy scene where the subject and background share colours — those need a trained segmentation model that understands objects, which a browser-side flood fill does not. Being honest about that boundary is more useful than promising a clean cut-out on every image and delivering a ragged one.

Setting the photo up for a clean cut

The matte only reaches background that is connected to the border, so the single biggest quality lever is the source photo, not the sliders. Frame the subject so the backdrop runs unbroken to all four edges, keep the background flat and evenly lit, and aim for a clear colour difference between subject and backdrop — a dark jacket on a pale wall separates far better than a pale one on white. A soft shadow cast onto the backdrop reads as "close to background" and gets eaten at higher tolerance, so diffused light and distance from the wall pay off before you ever open this page.

Finishing the cut-out

Once the alpha is set, the transparent PNG is raw material rather than a final asset. Drop it onto a new background and, if you need it small for the web, run it through the PNG to WebP Converter — WebP keeps the transparency at a fraction of the weight. If the cut is slightly low-resolution or you are enlarging the subject, the Image Upscaler sharpens it before compositing, and a Circle Crop turns the finished cut-out into an avatar in one step.

Transparent output, entirely on your device

The result is a PNG with a real alpha channel, previewed on a checkerboard so you can see exactly what was removed before you download it. It drops straight into Figma, Photoshop, Canva, a marketplace listing or a slide, and you can composite it over any new background. The whole process — decode, flood fill, feather, encode — runs on a local canvas. Nothing is uploaded, no account is required and no watermark is added, so an unreleased product shot or a personal photograph never leaves the machine it came from.