Background Remover

By RECATOOLS Editorial Updated 9 Aug 2026

About this tool

Removes a flat background in your browser, and tells you before you start whether your picture is one it can actually handle.

Remove an image background in your browser — nothing is uploaded. Works on flat backgrounds, and tells you before you start whether yours qualifies.

RT-IMG-030 · Image & File

Drop an image here or click to browse

PNG, JPG, WebP · nothing is uploaded

How to use the background remover

Drop in your image

It is read straight into the page. Nothing is sent anywhere, so a product shot you have not launched yet is as safe here as on your own disk.

Read the verdict before you touch anything

The tool measures how uniform the border of your picture is and says whether this technique suits it. A flat studio backdrop scores near 100%. A photo taken outdoors will score low, and the honest answer is to use something else.

Raise the tolerance until the background goes

Then stop. Too much tolerance starts eating the subject, and the first thing to disappear is whichever part of it is closest in colour to the background.

Download as PNG

PNG, not JPEG, because JPEG has no alpha channel. Saving a cut-out as JPEG silently fills the transparency with black.

Flood fill, not a neural network — and when that is the wrong choice

Almost every free background remover is a neural segmentation model running on somebody else's server. That is a genuinely better technique for hard pictures — it can separate hair from sky, which no amount of colour matching will ever do. It also means uploading your image to a third party, waiting on their queue, and accepting whatever they do with the file afterwards.

This tool takes the other trade. It runs a flood fill in your browser: it samples the border of the image, works out the background colour, and removes every pixel connected to the edge that is within a tolerance of that colour. Nothing is uploaded, there is no queue, and the result appears as fast as your machine can walk the pixels.

On a flat studio background this is not merely adequate — it is exact, because there is a right answer and the fill finds it. On a photo taken in a park there is no right answer to find.

Why it starts by judging your picture

The failure mode of colour-based removal is not that it breaks loudly. It is that it returns something plausible-looking with holes in the subject and fringes of background still attached, and the user spends ten minutes dragging a slider before concluding the tool is bad. So this one measures the border first and tells you what it found. If the border is one flat colour, you will get a clean cut. If it varies, you will be told that a flood fill is the wrong instrument before you have wasted any time on it.

The two rules that decide whether it works

First, the background must touch the edge of the frame. The fill starts from the border and spreads inward, which is why an enclosed region — the gap inside a handle, the hole in a doughnut — keeps its background even though it is the same colour. That is usually what you want, and when it is not, the fix is to crop rather than to raise the tolerance.

Second, the subject must differ from the background by more than the tolerance. A white mug on a white sweep is the hard case: there is no threshold that removes one and keeps the other, because at the pixel level they are the same colour. Photographers solve this with a coloured backdrop or a rim light rather than with software, and that remains the right answer.

Product photos, and why this matters in Southeast Asia

Shopee, Lazada and TikTok Shop all reward listings with clean, consistent product images, and small sellers across Singapore, Malaysia and Indonesia shoot against a sheet of white card precisely because it is cheap and it works. That is exactly the picture this technique is built for: flat backdrop, clear subject, no hair. A seller with fifty items to list does not want to upload fifty photographs to a service with a daily free-tier cap, and does not need a neural network to do what a threshold does correctly.

It is also the case that a product photo often contains information a seller would rather not hand over — pricing sheets in shot, an unreleased design, a supplier label. Local processing removes that question entirely rather than answering it with a privacy policy.

10 Things to Know About Removing Backgrounds

JPEG cannot hold transparency. There is no alpha channel in the format. Save a cut-out as JPEG and the transparent area is filled — usually with black — which is the single commonest way a good cut-out is ruined at the last step.

The fill is 4-connected, not 8. It spreads up, down, left and right but never diagonally. A one-pixel diagonal gap in an outline is therefore not a leak, which is what stops the background bleeding into a subject through a hairline.

Enclosed regions survive on purpose. The hole in a doughnut is the same colour as the background but is not connected to the border, so it stays. That is the behavioural difference between a flood fill and a colour key.

Colour distance here is Euclidean in RGB. That is not how eyes work — perceptually uniform spaces like CIELAB exist for exactly this reason — but it is predictable, and a person dragging a slider is doing the perceptual correction themselves.

Green screens are green for a reason. Human skin contains very little green, so the gap between subject and background is at its widest in that channel. Blue screens came first and are still used where the subject wears green.

A recursive flood fill overflows the stack. On a flat region a few hundred pixels across, the naive implementation dies. Every serious version is iterative with an explicit stack, including this one.

Feathering is why cut-outs stop looking cut out. A hard alpha boundary reads as scissors. Giving boundary pixels partial transparency lets the subject sit on a new background without a bright halo tracing its outline.

JPEG compression makes flat backgrounds less flat. A white sweep saved at a low quality acquires faint blocky variation, which is why a JPEG sometimes needs more tolerance than the PNG of the same shot.

Hair is the boundary of the technique. A strand thinner than a pixel is genuinely part background and part hair in that pixel, so no threshold can classify it correctly. This is precisely what segmentation models are for.

This one never uploads anything. The image is read into a canvas in your tab and the pixels are walked there. Open DevTools and watch the network panel while you use it: nothing goes out.

Frequently Asked Questions

No. The file is read into a canvas inside your browser tab and every pixel is processed there. Nothing is sent to us or to anyone else, and you can verify that in DevTools — the network panel stays silent while the tool runs. That is the main reason to use a local technique instead of a hosted model.
Because it measured the border of your picture and found it varies too much for there to be a single background colour. This technique removes pixels that match the background and are connected to the edge; if the edge is a garden, a street or a patterned wall, there is nothing consistent to match. Saying so up front is more useful than handing back a ragged cut-out.
The tolerance is too high, and the first part of a subject to vanish is whichever part is closest in colour to the background. Lower the tolerance until the subject is intact, and accept a little background residue — that is easier to fix than a hole in the middle of a product.
Because the fill spreads from the edge of the frame inward, and an enclosed area — the space inside a handle, for instance — is not connected to the edge. This is deliberate: it is what stops the tool eating background-coloured areas of your subject. Crop tighter, or remove that region separately.
Not well, and this is the honest limit of the method rather than a bug. A pixel containing one strand of hair is part hair and part background, so no single threshold can classify it. Segmentation models handle this because they predict a soft mask rather than applying a rule. If your picture is a portrait, use one of those.
JPEG has no alpha channel — transparency simply does not exist in the format. Exporting a cut-out as JPEG fills the transparent region, normally with black, and undoes the work. PNG and WebP both carry alpha; the download button here produces a PNG.
The default of 12 suits a clean studio background. A JPEG of the same shot often needs more, because compression introduces faint variation into what was a flat colour. Raise it in small steps and stop the moment the background has gone rather than pushing for a rounder number.
It samples every pixel around the four edges of your image, takes their mean colour, and reports how far the border pixels sit from that mean on average. A perfectly flat backdrop scores close to 100%. It is a prediction about whether a single background colour exists at all, made before any pixels are changed.
No, and nothing threshold-based will. At the pixel level the subject and the background are the same colour, so there is no value that separates them. This is solved when the photograph is taken — a coloured backdrop or a rim light — rather than afterwards in software.
They work, but the fill walks every pixel, so a 6,000-pixel-wide photograph takes noticeably longer than a web-sized one. If the cut-out is destined for a web page, resizing before removing the background is faster and produces the file you actually needed.

Related News

You may be interested in these recent stories from our newsroom.

View all news →

Background Remover