Runs on your device · WebGPU

Remove the background from a photo

Add a photo and an AI model running on your own device cuts out the person, product or pet. Keep it see-through as a PNG, or set it on white, a colour of your choice or a blurred copy of the photo, then save it at full size. Up to 10 photos at a time, and none of them is uploaded.

Downloads once: BiRefNet-lite (98 MB) for any subject, or MODNet (26 MB) for people, plus a runtime of 14 to 27 MB

How it works

  1. 1Add up to 10 photos

    JPEG, PNG or WebP, up to 40 MB each, or paste a screenshot. Pictures above 16 megapixels are scaled down to 16 MP as they open.

  2. 2Pick a model

    Best (BiRefNet-lite) finds any subject: people, products, pets, drawings. Fast (MODNet) is trained on portraits and is about a quarter of the download.

  3. 3Remove the background

    The first run downloads the model and your browser keeps it. After that a photo takes a few seconds on a graphics chip, and several times longer on the processor.

  4. 4Choose a backdrop and save

    Drag across the picture to compare before and after. Save each photo at its full size, or all of them in one ZIP.

Best or Fast: which model to use

Model choiceBest (any subject)Fast (people)
Model and licenceBiRefNet-lite, MIT licenceMODNet, Apache 2.0 licence
FindsWhatever stands out: people, products, pets, logos, drawingsPeople only; objects on their own are ignored
One-time download98 MB, plus a 14 to 27 MB runtime26 MB, plus a 22 MB runtime
EdgesSharp outlines, good on products and flat graphicsSoft, natural hair

Best treats everything that stands out as the subject, so a hand holding a bottle stays in the cut-out, and so does a second person at the edge of the frame. Crop the photo first if you only want one thing. Fast looks for people and nothing else: a product shot gives it nothing to find, while for a selfie or a group photo it is the quicker, lighter choice. If Best finds no clear subject in a photo of a person, the tool offers to try Fast.

Clean edges on a new background

The model looks at your photo at about 512 pixels across and works out, for every pixel, how much of it belongs to the subject. That map is then stretched back over the full-size original with a soft edge, so the saved file keeps every pixel of detail inside the subject. The Edge softness slider sets how that edge behaves: Crisp tightens it, which suits products and logos; Natural leaves it as the model drew it; Soft feathers it, which can hide a rough outline.

Hair and fuzzy edges are part subject and part old background. Left as they are, a dark room behind someone shows as a grey fringe once they sit on white. The tool replaces the colour in those see-through pixels with the subject’s own nearby colours before the new background goes in, which is what removes that fringe.

  • Blurred photo puts the subject back on a softened copy of its own photo, with the subject painted out first so no glow shows around it.
  • Transparent saves as PNG. White and colour backgrounds can be JPEG, which is far smaller, or PNG.
  • A picture that already has see-through areas, such as a logo or a cartoon avatar, keeps them in the result.

What runs where

Everything happens inside this browser tab. Your photos are opened, cut out and saved by your own device, and nothing about them is sent to us or anyone else. The only things downloaded are the model, from Hugging Face, and the runtime that runs it, from the jsDelivr network, once each; after that the browser keeps them.

The work goes to your graphics chip through WebGPU where the browser offers it. According to web.dev (November 2025) that includes Chrome and Edge on computers, Chrome on many Android phones, Safari on iOS 26 and macOS Tahoe, and Firefox on Windows. Elsewhere it runs on the processor, which works the same but is slower. Every result is checked: if a graphics chip returns an empty or broken mask, the photo is run again on the processor and the chip is skipped for the rest of the visit.

Tip: Cutting out a product for a shop? The product photo maker centres it on a clean background in a square frame, the shape Meta’s catalogue image guidance recommends. For a round profile photo, the profile picture maker frames your head and shoulders.

Questions people ask

Is this background remover free?

Yes, with no sign-up, no watermark and no daily limit. You can work on 10 photos at a time and start a new batch as often as you like. It can be free because your own device does the work.

Are my photos uploaded anywhere?

No. They are opened, cut out and saved inside your browser. The model and its runtime are downloaded to your browser once; your photos never travel the other way.

Why is the first photo slow?

The first run downloads the model: about 125 MB for Best on a computer with WebGPU, or about 48 MB for Fast. Your browser keeps it, so later visits start at once. Without WebGPU every photo also takes longer, because the processor does the work.

Why did it keep a hand or a second person?

Best keeps whatever stands out in the photo, and a hand holding a product usually does. Crop the photo to what you want before adding it. For photos of people, Fast (people) is the better pick; it ignores objects.

What size is the saved file?

The same size as your photo, up to 16 megapixels (bigger photos are scaled down to that when they open). Transparent cut-outs are PNG; white, colour and blurred backgrounds can be JPEG or PNG.

Does it work on a phone?

Yes. Phones with WebGPU, such as iPhones on iOS 26 and many Android phones in Chrome, run it on the graphics chip. Others use the processor and take longer. The download is large, so use Wi-Fi the first time.

More free tools

Related guides

Answer every comment and DM, automatically

When someone comments a word like GUIDE, DMFast replies and sends them your link in a DM. Free plan, no card.

Start free