Remove the background from a photo

Getting a clean cut-out

How it works

An AI segmentation model looks at your photo and decides, for every pixel, how likely it is to belong to the main subject. That gives a mask, and the mask becomes the transparency of the PNG. Everything runs inside your browser — the model files are downloaded once and then reused from your browser cache.

Portraits or Any subject?

Any subject (BiRefNet-lite) is the default: it handles products, pets, plants, everyday objects and people, and keeps fine edges such as fur and straps. Portraits (MODNet) is much smaller (about 7 MB) and fast, but it only works for people and faces — on anything else it can leave holes or remove the whole subject.

Photos that work best

A clear subject, good lighting and some contrast with the background give the cleanest result. Busy backgrounds, motion blur, glass, smoke and hair against a similar color are harder. When the first result is not perfect, switch model or use the touch-up brush.

Touch up with the brush

Erase removes leftovers from the background; Restore brings back parts the model cut away. Zoomed-out strokes are replayed at full resolution on export, so what you see is what you download.

PNG, JPG and backgrounds

A transparent PNG is best for stickers, product listings and overlays. If you pick a white, black or colored background, the export becomes a JPG on that color — handy for marketplace photos that require a plain background.

Licensing and privacy

The models used are open source (MODNet: Apache-2.0, BiRefNet-lite: MIT). Your photo is never uploaded to any server. Only the model files are fetched from public servers the first time. You are responsible for having the rights to the images you edit.