Removing the background from a photo: what works and what does not

You need a product on a white background for a listing, or a face cut out for a team page. Background removal has gone from a job for a designer to something that takes a few seconds, but the results still swing wildly between perfect and unusable.

The difference is almost never the tool. It is the photo.

What the tool is looking for

Background removal works out which pixels belong to the subject and which do not. It is good at the things people are good at: a clear outline, a subject that differs from what is behind it, even lighting.

It struggles with exactly what you would struggle with if you were tracing the edge by hand:

• Hair and fur. Thousands of thin strands, each partly transparent against the background. This is the hardest case in the whole field.
• A subject the same colour as the wall. A grey jumper against a grey wall has no edge to find.
• Glass and anything transparent. A wine glass is mostly background. Should it be cut out? There is no correct answer, so the result is a guess.
• Motion blur. A blurred edge is not an edge.
• Deep shadows. A shadow attached to the subject reads as part of it, and often survives the cut.

Shooting so it works

Two minutes of care beats an hour of fixing:

1. Put distance between subject and background. Half a metre is enough. It also softens the background and gives a cleaner edge.
2. Contrast, not matching. Dark object, light wall. Light object, dark wall.
3. Even light, no hard shadow. A window to the side beats a ceiling light overhead. Overcast daylight is the friendliest light there is.
4. Plain background. A blank wall or a sheet of paper. Patterns give the algorithm ideas.
5. Shoot larger than you need. Detail helps, and you can always scale down afterwards with Resize image.

Doing it

Open Background remover, drop the photo in and wait a few seconds. You get a PNG with a transparent background.

A word about what happens to your picture. This one is different from most tools here: the model that does the separation runs on the server, so the image is sent, processed and then deleted. It is not kept and it is not used to train anything. If that is not acceptable for a particular photo, do not use it for that photo.

PNG is the right output format because it can store transparency. If you save the result as JPG the transparency is lost and comes back as white, which is fine when you are placing it on white anyway and wrong everywhere else.

Fixing what it gets wrong

No automatic cut is perfect on a hard photo. For the last few percent, open the result in Image editor:

• The eraser cleans up leftover background at the edges.
• Layers let you drop your own background behind the cutout and see the join.
• Slight blur along the edge hides the stair-stepping that a hard cut leaves.

For marketplace listings, an honest tip: a plain white background sells better than a clumsy cutout. If the edge looks chewed, put the subject on white and leave a natural shadow rather than fighting for a perfect silhouette.

Frequently asked questions

Why did it cut off part of my subject?

Usually low contrast between that part and the background, or a strong shadow. Photograph it again against a contrasting background, or repair the edge in the image editor.

Can it handle hair properly?

Partly. Modern models do a decent job on a clear portrait with a plain background, but wispy strands against a busy scene still lose detail. It is the hardest case there is.

What format should I save the result in?

PNG, because it keeps transparency. WebP also supports it. JPG does not: it fills transparent areas with white.

Is my photo uploaded?

Yes, this one runs on the server, because the model is too large to send to a browser. The picture is deleted after processing and is not used for training. Most of the other tools on the site never upload anything.

Are there limits?

Free accounts get a number of runs per day, Premium gets more and accepts larger files. The exact figures are on the pricing page.