RunToolRunBackground Remover (AI)
Automatically remove backgrounds with AI.
How to Remove a Background with AI
Select a photo
Click the upload area or drop a JPG, PNG or WebP image with a clear subject.
AI processes it on your device
A neural network downloads once, then detects the subject and erases the background — entirely in your browser.
Download the transparent PNG
Save the result with a fully transparent background, ready for design work or product listings.
How the cutout is decided
Background removal here is a segmentation model: a neural network trained on a very large number of images to predict, for every pixel, how likely it is to belong to the foreground subject. The output is not a binary mask but a matte of values between zero and one, which is what allows soft edges instead of a cut-out-with-scissors look.
The model runs on your device. Where the browser supports WebGPU it uses the graphics hardware and completes in a few seconds; otherwise it falls back to a slower path. Either way the image is never uploaded, which is the main practical difference between this and the well-known services that require an account.
What it handles well and what defeats it
Clear subject–background separation is the easy case: a person or product against a contrasting, reasonably uniform backdrop produces an excellent matte with almost no effort. Product photography on white is close to a solved problem.
The hard cases are all about ambiguity. Fine hair against a busy background, transparent and semi-transparent objects like glass or veils, motion blur, and subjects whose colour closely matches what is behind them. A model predicting per-pixel membership genuinely cannot tell where a strand of hair ends and a similarly coloured wall begins, because the pixel contains both.
Edge quality and colour fringing
Where the original background was strongly coloured, its colour bleeds into the semi-transparent pixels along the subject's outline. Placing that cutout on a new background leaves a visible rim of the old one — the green fringe familiar from badly keyed video. It is most obvious when moving from a dark background to a light one or the reverse.
This is a property of the source image rather than a flaw in the removal. Shooting against a background whose colour is unlike the subject, with even lighting and a little physical distance between subject and backdrop, prevents most of it. Fixing it afterwards means shrinking the matte slightly or decontaminating the edge colour, which is retouching work rather than a setting.
Choosing what goes behind
A transparent PNG is the flexible output: it keeps the alpha channel so the subject can be placed on anything later. Saving as JPEG destroys that immediately, because JPEG has no transparency and fills it — usually with black. If the result is going into a design tool, keep it PNG or WebP.
Replacing the background with a solid colour is what most marketplace listings actually require, and matching that colour to the destination platform's own background is what makes a product look like it belongs there. A photographic replacement background is harder to make convincing: lighting direction and colour temperature have to match, and when they do not, the composite reads as fake even to viewers who cannot explain why.
Resolution and expectations
Segmentation runs at a fixed internal resolution and the matte is scaled to your image, so extremely large photographs gain less than you might expect at the edges. Feeding in a reasonably sized image often gives a cleaner-looking result than feeding in a 50-megapixel original.
For a marketplace listing, a social post or a presentation, an automatic cutout is genuinely sufficient. For print at large size, or for a subject with complicated hair against a complicated background, expect this to be the first step of a manual process rather than the whole job.