Scrawlspace › Image Tools

Remove Background from Image

Cut a person or object out of a photo and save it as a transparent PNG, or put it on a white, light blue, custom colour or blurred background. The photo is never uploaded: an AI model downloads once and then runs on your device, so it suits ID and passport photos you'd rather not send anywhere.

How to use Remove Background

  1. Drop one or more photos here, paste one with Ctrl+V, or choose files. JPG, PNG, WebP and iPhone HEIC photos all work.
  2. Leave Subject on Person for portraits and ID photos. For products, pets and objects choose Any subject.
  3. Pick a background: None for a transparent PNG, White for passport and exam photos, Blue, your own Colour, or Blur to keep a soft-focus copy of the original behind the subject.
  4. Drag across the preview to compare before and after. If the outline looks too sharp, raise Soften edge.
  5. Download the photo, or all of them as a .zip. Then size it with the passport size photo maker or compress it to 50 KB if a form needs that.

Frequently asked questions

Is my photo uploaded anywhere?

No. The photo is read and processed in your browser and never leaves your device. What's downloaded is the AI model itself, from Hugging Face, and the engine that runs it, from the jsDelivr CDN. Nothing about your photo is sent with those downloads.

How big is the download, and does it happen every time?

Person mode downloads a model of about 26 MB and Any subject one of about 5 MB, each the first time you use it. The engine that runs them adds about 6 MB. The models are kept in your browser's storage and the engine in its cache, so later visits start without downloading them again. Clearing this site's data removes the models.

How do I make a white background for a passport or exam photo?

Choose Background: White and Save as: JPG, then download. The tool only replaces the background; it doesn't crop or resize. Next, use the passport size photo maker for 35×45 mm or 2×2 in, and check your form's limits in the exam photo and signature size guide.

How do I get a transparent PNG?

Choose Background: None. The result is always saved as a PNG, because JPG can't hold transparency. The checkerboard in the preview shows the see-through areas.

Which photos work best?

A clear subject that stands out from its background: a person facing the camera, or a product on a plain surface. Hair is usually kept, but stray strands can be lost or keep a little of the old background around them. Glass, smoke, see-through fabric and subjects the same colour as the background are hard, and parts may be cut away or left in. Soften edge hides a hard outline but can't put back what the model missed.

Person or Any subject: which should I use?

Person uses MODNet, a model made for portraits: it keeps hair and soft edges best and suits passport, ID and profile photos, but it only looks for people. Any subject uses U²-Netp, a small model that picks out the main object in a photo, such as a product, pet or vehicle. Its edges are rougher, especially around hair, so use Person whenever the subject is a person. If Person finds nothing in a photo, the tool suggests switching.

Does it work on a phone?

Yes. Both modes work in mobile browsers; expect a few seconds per photo, more on older phones. If the phone runs out of memory the tool says so, and closing other tabs or using a smaller photo usually helps. Photos larger than 8,000 pixels on a side or 40 megapixels are scaled down to fit.

Is it free? Is there a watermark?

It's free, with no sign-up and no watermark, and you can add as many photos as you like. The output is the same size as your photo, up to the 40 megapixel limit.

Backgrounds and formats

BackgroundSaved asGood for
NonePNG with transparencyLogos, stickers, product listings, designs and slides
WhiteJPG or PNGPassport, visa and exam photos; see the exam photo size guide
Blue (light)JPG or PNGForms and ID cards that ask for a light blue background
ColourJPG or PNGAny colour you pick, or type as a hex code such as #f2f2f2
BlurJPG or PNGA portrait look that keeps the original scene, softly out of focus

How it works

A segmentation model looks at a reduced copy of your photo (512 pixels on the short side for Person, 320 × 320 for Any subject) and marks how much of each pixel belongs to the subject. That mask is then stretched back over the photo at its full size, so the output keeps your original resolution and detail.

It runs on your device's processor through WebAssembly, usually in a few seconds per photo. The page stays responsive while it works.

Person uses MODNet and Any subject uses U²-Netp, both open models under the Apache-2.0 licence, downloaded from Hugging Face the first time you use them.

Getting a clean cut-out