Skip to content
RoomCalcs

Remove background from image

An AI model cuts out the person, pet or product in your photo in seconds. Keep it transparent (PNG), put it on white or any colour, blur the old background or use your own photo — then fix any spot with the Erase and Restore brushes and download at full resolution. It all runs in your browser: your photo is never uploaded.

  • Free
  • No sign-up
  • Private — runs in your browser

How to remove the background from a photo

  1. 1Choose a photo — JPG, PNG, WebP, or HEIC where your browser supports it.
  2. 2Wait a few seconds while the AI removes the background (the first time it also downloads the model, once).
  3. 3Pick a new background: transparent, a colour like white or ID-photo blue, a blur of the original, or your own photo. Check the edges with the before / after slider.
  4. 4Touch up with Erase or Restore if needed, crop to the subject if you like, and download as PNG, JPG or WebP.

How does the AI background remover work?

  1. 01AI mask: MODNet (Quick, at 512 pixels on the short side), IS-Net (Standard) or BiRefNet lite (Best) looks at the photo at 1024 × 1024 and scores every pixel 0–1 "subject"
  2. 02Full-size edge: guided filter — alpha = a·(R, G, B) + b, fitted near the edge to the full-resolution photo
  3. 03Clean-up: below 6 % → transparent, above 80 % → solid, smooth curve in between
  4. 04Edge colour: F = F̄ + α(I − αF̄ − (1 − α)B̄) (blur-fusion foreground estimate)
  5. 05New picture = F × α + background × (1 − α)

The AI model is a neural network trained on thousands of photos where the subject had been cut out by hand. It looks at a 1024 × 1024 copy of your photo and gives every pixel a score between 0 (background) and 1 (subject). That score becomes the alpha channel — how see-through each pixel is.

A 1024-pixel mask is too coarse for a 12-megapixel photo, so the tool doesn't just stretch it. Near the outline it fits a guided filter to your full-size photo: inside each small window the mask is modelled as a straight-line function of the red, green and blue values, so the edge follows the real colour boundary in the photo instead of the blurry, scaled-up mask. Areas that are clearly subject or clearly background keep the model's answer.

Then a smoothstep curve clears the faint haze the model leaves in the background (under 6 %) and makes the subject fully solid (over 80 %). Pixels on the edge — hair, fur, motion blur — stay partly transparent. Their colour is still a mix of subject and old background, which shows as a halo on a new background, so the tool estimates the pure subject colour F from nearby solid subject pixels and the background colour B from nearby background, and solves the mixing equation I = αF + (1 − α)B for F.

Be honest with yourself about hard cases: fine hair and fur against a busy background, glass and smoke, and a subject whose colour matches the background (a white shirt on a white wall) are harder for any AI. That is what the Erase and Restore brushes are for — paint over the spot and only that area changes.

Good to know

  • For a passport or visa photo, choose a white background (or the colour your form asks for), then use our passport photo maker to get the exact size.
  • Photos with good light and a clear gap between the subject and the background give the cleanest edges. Hair against a plain wall works much better than hair against trees.
  • Product photos for online shops: use a white or light-grey background, turn on "Crop to the subject" with a little space, and save as JPG.
  • If the cut-out has a thin halo of the old background, keep "Remove colour fringe" on and set Edge softness to 0–10; use the Erase brush for anything left over.

Background remover — common questions

Is this background remover really free, with no watermark?

Yes. There is no sign-up, no watermark and no limit on size — you download the picture at the full resolution of your photo (very large photos over 25 megapixels are scaled down to fit your device's memory, and you'll see a note).

Is my photo uploaded to a server?

No. The AI model is downloaded to your browser the first time (from this site, in small pieces that resume if the connection drops) and saved there, and then everything — the background removal, touch-ups and saving — happens on your own device. Your photo never leaves it.

How do I make the background transparent?

Keep the background on Transparent (the checkerboard) and download as PNG or WebP. JPG can't store transparency, so a JPG gets a white or coloured background instead.

How do I change the background colour of a photo to white or blue?

After the background is removed, choose Colour and tap White, ID photo blue or any other colour — or type a hex code. Then download as JPG or PNG.

Why are hair or some edges not perfect?

Fine hair, fur, see-through objects and a subject that has the same colour as the background are the hardest cases for any AI. Try Best quality if your browser supports it, keep "Follow the photo's real edges" on, and use the Erase or Restore brush on the spot that's wrong.

What is the difference between Quick, Standard and Best quality?

Quick uses the small MODNet model (7 MB): the smallest download and a result in seconds, made for photos of people — it is chosen automatically on phones and slow connections. Standard uses the IS-Net model (44 MB) for people, pets and products with sharper edges; it runs on every device but takes about half a minute on a computer and a minute or more on a phone. Best quality uses the larger BiRefNet model (115 MB) for finer edges and needs a browser with WebGPU, such as Chrome or Edge on a computer or a recent Android phone. If WebGPU isn't available, Standard is used automatically.

Sources

Formulas last reviewed: September 30, 2026 ·How we check our formulas

Related tools