How to make a white background transparent (and when not to)

Removing a white background from a logo takes about five seconds if the background really is flat white, and produces a crunchy mess if it is a photograph of a white wall. Knowing which situation you are in is most of the skill.

What colour keying actually does

Our colour removal tool works by comparison, not by understanding. You pick a colour; it walks every pixel, measures how far that pixel is from your chosen colour, and makes it transparent if the distance is under your tolerance.

There is no notion of a subject or a background involved. If your logo contains white letters and the background is also white, both disappear — the tool has no way to know you meant only one of them. That is the fundamental limit, and it explains most disappointing results.

We measure distance with a weighted formula that counts green most heavily, then red, then blue, approximately matching how sensitive human vision is to each. It means the tolerance behaves more predictably than a raw numeric comparison would.

Choosing a tolerance

Tolerance is the one setting that matters. It runs from 0 to 100 and controls how different a pixel may be from your picked colour while still being removed.

Start low and climb

0–5 removes only near-exact matches. This is right for graphics exported straight from design software, where the background is mathematically identical everywhere.

10–25 is the useful range for most real files. It absorbs the small variations introduced by JPEG compression, which never reproduces a flat colour perfectly. If you loaded a JPEG logo, start around 20.

30–50 handles scans and photographed documents, where paper is unevenly lit. Watch the subject carefully in this range — light areas inside the subject start to vanish.

Above 60 is rarely what you want. At high tolerance the distance threshold is wide enough to swallow large parts of the subject.

The preview updates as you drag, so the reliable method is to raise the tolerance slowly until the background is gone, then stop immediately. The last few points before the subject starts eroding is the sweet spot.

Reading a bad result

A pale halo around the subject

Almost always means tolerance is slightly too low. Anti-aliased edges are blends between the subject and the background — a letter’s edge pixel might be 60% white. Those partial pixels sit just outside your threshold and survive as a fringe. Nudge tolerance up a few points.

Our tool feathers the boundary rather than making a hard cut: pixels just beyond the threshold get partial transparency instead of staying fully opaque. That is why edges come out reasonably smooth rather than jagged, but it cannot rescue a tolerance that is far too low.

Holes in the subject

Tolerance is too high, or the colour you picked genuinely appears in the subject. If a product photo has white highlights, keying out white will punch holes through them. There is no tolerance value that fixes this — the tool cannot distinguish the two whites.

Nothing happens

Usually the picked colour is not what you think. A “white” background is often #FDFDFB or similar. Click directly on the background in the preview to sample the real value rather than assuming pure white, or use the eyedropper.

Save as PNG, not JPEG

This trips people up constantly: you remove the background, export as JPEG, and the white is back. JPEG has no alpha channel, so transparency cannot survive it — the transparent areas get flattened onto white on the way out.

Export as PNG, or WebP if you want a smaller file. In our format measurements, WebP was substantially smaller than PNG on a transparent graphic while preserving the alpha channel fully.

When to use something else

Photographic backgrounds — a person against a room, a product on a textured surface — need subject segmentation, not colour matching. A machine-learning background remover understands “this is a person”; colour keying only understands “this pixel is close to that colour”. We deliberately do not ship an AI remover, so this is a case where another tool is the right answer.

Fine detail like hair or fur defeats colour keying entirely. Individual strands are blends of subject and background across their whole width.

If you have the original vector file — an SVG or AI logo — export a transparent PNG from that instead. It will be perfect, and no keying is involved.

Doing a batch

For a set of logos or scans sharing the same background, bulk colour removal applies one colour and tolerance across every file and returns transparent PNGs as a ZIP. Tolerance absorbs small differences between files, but if the backgrounds are genuinely different colours, process them in separate batches.

If some images need individual attention, the editor lets you set a batch-wide removal and then adjust a single image on top of it, stepping through them with the arrows.