Remove the Background from Any Image

The background remover separates the main subject of a photo from whatever surrounds it. It uses a salient-object segmentation model, a neural network trained to estimate, for every single pixel, how likely that pixel is to belong to the thing a viewer would look at first. That estimate becomes the alpha channel of a new PNG: confident subject pixels stay fully opaque, clear background pixels turn fully transparent, and the soft band along the edge keeps partial transparency so the cutout does not look stamped out with scissors.

Results are always delivered as PNG because a cutout is only useful if the empty area stays empty. JPG has no alpha channel at all, so a transparent region saved as JPG would simply be painted a solid colour. PNG keeps the transparency losslessly, which makes it the dependable choice for slides, storefronts and design tools. If the file is too heavy for a web page, run it through the image converter and choose WebP, which also supports transparency at a much smaller size.

The model does best when the subject is clearly distinct from its surroundings: a person against a wall, a shoe on a table, a dog on grass. It finds harder material more challenging. Wispy hair, glass, smoke, mesh fabric and very busy scenes with several competing subjects can leave a halo or remove a little too much. The prediction is made at 1024 by 1024 pixels and then stretched to the size of your photo, so extremely fine strands may soften slightly on a large original.

A few limits keep the tab quick and stable. Bulk mode holds 25 images, each file can be up to 15 MB, and anything larger than 16 megapixels is scaled down proportionally before the cutout is drawn. Images are processed one after another in a separate thread so scrolling and typing stay smooth. The first image takes longer because the roughly 44 MB model and 14 MB runtime have to download; your browser keeps both for the next visit.