How to Remove the Background from an Image

At some point almost everyone doing anything online needs to remove background from an image. You're listing a product on a marketplace that requires a white background. You're cutting a logo out of a screenshot so it sits cleanly on a slide. You're prepping a headshot for a team page and the wall behind it is distracting. The task is common enough that entire companies exist to solve it, and most of them want either a subscription or a credit card before they'll hand you a finished file.

You don't always need that. If your background is a single, mostly even color (a studio backdrop, a solid wall, a flat-color logo canvas) you can get a transparent background image online in a few seconds, for free, without uploading anything to a server. This post covers when that approach works, when it doesn't, and how to get a clean result out of it.

Two very different ways to remove a background

There are really two families of background removal tool, and mixing them up is where most disappointment comes from. The first is color-based removal, sometimes called chroma keying: the same principle a weather forecaster's green screen relies on. You tell the software (or it guesses) which color counts as "background," and it makes every pixel close enough to that color transparent. It doesn't know what a person or a shoe or a bottle is. It only knows colors and how close they are to each other numerically.

The second is AI-based subject detection, the kind of thing remove.bg and similar paid tools do. These run an image through a trained segmentation model that's seen millions of photos and learned to guess where a "subject" ends and a "background" begins, regardless of what color either one is. That's genuinely useful, and it's the only realistic way to cut a person out of a busy street photo, but it takes real compute, which is why those services meter you with credits or a subscription.

GlaeKit's Background Remover is the first kind. It samples a color, automatically from the corner pixel or one you pick manually, and removes every pixel within a tolerance range of that color. That's the whole mechanism. There's no model, no server call, no upload. It runs in JavaScript in your tab, which is also why it's instant and why nothing ever leaves your device.

Neither approach is "better" in the abstract. For a product photo on a seamless white sweep, color-based removal gives you a perfect edge in under a second, something AI tools sometimes actually struggle to do cleanly because they're busy guessing at object boundaries that were never ambiguous in the first place. For a photo of someone standing in a park, color-based removal is close to useless. The grass, their skin, and their shirt might all fall into overlapping color ranges, and there's no single "background color" to key off of.

Getting a clean result from color-based removal

If you want to remove image background free and get something usable, the photo matters more than the tool. A handful of habits make the difference between a crisp cutout and a ragged one.

Light your background evenly. A single soft light source cast across a plain backdrop will still produce a subtle gradient, brighter near the light and dimmer at the edges, and that gradient is a range of slightly different colors, not one color. Tighten the tolerance and you'll leave a faint halo of that gradient behind; loosen it and you start eating into your subject. Two diffuse lights, one from each side, flatten that out considerably.

Watch for shadows. A subject sitting close to its backdrop throws a soft shadow onto it, and that shadow is a darker shade of the same background color, sometimes different enough to survive removal, sometimes not, depending on tolerance. Pulling the subject a foot or two forward, away from the backdrop, thins the shadow out and usually solves this without touching any settings.

Contrast is the other lever. If your subject shares a color family with the background (think a white product on an off-white table, or dark hair against a dark backdrop) there's no tolerance setting that cleanly separates the two, because to the algorithm they're not clearly two different things. Swapping to a backdrop that contrasts with your subject (a green product on white, not on light gray) fixes more cutout problems than any amount of slider tweaking.

One specific quirk worth knowing: tolerance in this tool is measured as a numeric distance between colors, not a percentage, and the default of 40 is tuned for backdrops with a bit of lighting variation. On a genuinely flat, evenly lit studio background you can often drop it to somewhere around 15–20 and get a tighter edge with zero bleed into the subject — worth trying before you assume the photo itself is the problem.

What "transparent PNG" actually means

People say "transparent background" loosely, but the technical piece behind it is specific: an alpha channel. A standard RGB image stores three numbers per pixel: red, green, blue. A PNG with transparency stores a fourth number per pixel, alpha, which controls opacity from fully see-through to fully opaque. When this tool "removes" a background pixel, it isn't deleting anything or painting it white — it's setting that pixel's alpha value to zero. The pixel is still there, still holds a color, it's just invisible when the image is composited on top of something else.

This is exactly why the output has to be a PNG and can't be a JPEG. JPEG's compression format has no alpha channel at all. It was designed decades ago purely for photographic color data, and every JPEG pixel is fully opaque by definition. Run a "background removed" image through a JPEG re-save and the transparency simply vanishes; whatever background color your viewer defaults to (usually white) fills back in. PNG, and formats like WebP that support alpha, are the only sane targets if you actually need to keep the transparency for later use: dropping it onto a colored slide, layering it in a design tool, or using it as a favicon.

It's also worth knowing that alpha isn't binary. A pixel doesn't have to be fully opaque or fully transparent. It can be 50% see-through, which is exactly what lets a soft, feathered edge exist instead of a hard jagged one. Color-based tools generally do produce some partial transparency right at the boundary between kept and removed pixels, which is part of why edges usually look smoother than you'd expect from a "yes/no per pixel" description of the technique.

Try it

GlaeKit's Background Remover samples your background color automatically, lets you fine-tune tolerance with a live before/after preview, and exports a transparent PNG, all in your browser, nothing uploaded anywhere. It's built for solid or near-solid backgrounds: product shots, logos, flat-color scans. For a busy or gradient photo background, you'll want an AI cutout tool instead. This one will tell you honestly when it's the wrong fit.

Frequently asked questions