Why Your Background-Removed Photo Has a White Halo
The cutout looks right at a glance, then you drop it onto a dark background and notice a faint, thin white or light-gray line tracing the subject's outline, hair, shoulders, the edge of a product. It shows up even when the tool did its job correctly. That's not a mistake in the segmentation. It's an optical fact about how the original photo was captured, and it's fixable if you know where it comes from.
The edge pixels were never 100% subject
A camera sensor (and the image encoder afterward) doesn't produce perfectly sharp boundaries between a subject and its background. At the actual edge of any real-world object, the pixels sit exactly on the transition, and the sensor captures an optical blend of subject color and background color in that thin strip, a form of natural anti-aliasing baked into the original photo before any editing happens. If your subject was photographed against a white backdrop, that fringe of edge pixels is a mix of the subject's true color and white.
Background removal replaces the background with transparency, but it can only assign an alpha value per pixel, it can't retroactively un-mix a color that was already blended at capture time. Even a mathematically perfect alpha mask leaves those edge pixels holding their original blended color, which still has some of the old background's tint in it. Composite that cutout onto a new background and the old color reappears as a rim, because the pixel itself was never purely "subject" to begin with.
Why it's worse on high-contrast swaps
The halo is proportional to how different the old and new backgrounds are. A faint white fringe left over from a white studio backdrop is nearly invisible if you place the cutout on a light gray page. Place the exact same cutout on a black or saturated background and that same fringe suddenly reads as an obvious bright outline, because now it's the only light pixel in an otherwise dark region. The fringe didn't get worse. The contrast that reveals it did.
What actually reduces it
Two practical options, depending on whether you control the original photo:
- If you're shooting the source photo yourself, use a backdrop closer in tone to where the cutout will ultimately be placed. A neutral mid-gray backdrop fringes less visibly against both light and dark destinations than a pure white or pure black one.
- If you're stuck with a photo you didn't shoot, erode or feather the mask edge by a pixel or two before compositing. This trims away the thin ring of contaminated edge pixels entirely rather than trying to recolor them, at the cost of a very slightly tighter cutout.
There's no setting that removes this at the segmentation step, because segmentation isn't where the problem originates. The blended pixel exists in the source photo before any AI model touches it.
Using it
Remove Background runs entirely in your browser. For the segmentation mechanics behind the cutout itself, see How Does AI Background Removal Actually Work? For which formats can actually hold the transparent result, see which image file types support transparency.