pixelate.run

Blur part of an image without uploading it

Open a photo or screenshot. Faces and private text are found and blurred for you, and a dragged box blurs any other part. All of it happens in your browser.

Drop or paste an image herePress Ctrl+V (⌘V on a Mac), or click to choose a file.

A made-up order screen. Yellow marks what was found; the rest shows it covered.

How it works

  1. Open an imageChoose a file, drop it anywhere on the page, or paste it from the clipboard.
  2. Adjust what is blurredTap a blurred area to clear it, tap a line of text to blur it, or drag a box over anything else. The slider sets how strong the blur is.
  3. Save itDownload the result, copy it or share it. Location and camera data are left out of the copy.

What it looks for

Text is read with on-device recognition. On a computer, a privacy model then reads it as language, so names, nicknames and addresses are found even without a label in front of them; these patterns are checked as well. Faces and codes are found by looking at the picture itself.

  • Phone numbersUS, UK and European layouts, with or without a country code.
  • Email addresses
  • Street addressesEnglish, German, French, Spanish and Italian forms, UK postcodes, and US town, state and ZIP.
  • ID numbersUS social security, UK National Insurance and Chinese resident ID numbers.
  • Bank cardsConfirmed with the card’s check digit.
  • Bank accountsIBANs, verified by their check digits, and crypto wallet addresses.
  • Keys, tokens and passwordsAPI keys and access tokens by their prefixes, and whatever follows a “Password” label.
  • Order and tracking numbersAny long run of digits, with or without a letter prefix.
  • IP addressesIPv4, IPv6 and hardware (MAC) addresses.
  • Names, birth dates and other labelled detailsWhatever follows a label such as “Recipient”, “Date of birth” or “CVV”, in seven languages.
  • FacesPortraits and group photos.
  • Profile pictures and nicknamesIn screenshots of chats, contact lists and feeds.
  • QR codes and barcodes
  • Licence platesChinese plates, where they appear as text.

For anything it misses, tap a line of text to cover it, pick single words out of a line, or drag a box over any part of the image.

Seven ways to cover

  • BlockA solid colour. Nothing of the original is left underneath.
  • PixelateA mosaic. Text is shuffled first, so it cannot be reconstructed.
  • BlurA soft smear, also made from shuffled text.
  • MarkerA thick felt-tip stroke.
  • InkAn uneven patch, as if dabbed on by hand.
  • EmojiA row of emoji, or one large one over a face.
  • Hide textFilled with the surrounding background, so the text is simply gone.

Questions

Can a blurred image be unblurred?
A light blur over text can sometimes be reversed. Here, text is first broken into shuffled blocks and only then blurred, so the original cannot be worked back out. Blurred faces keep their rough shape, which is what makes them look natural; use a stronger setting or a solid block when identity really matters.
Can I blur the whole image?
Yes. Choose “Blur the whole image” in the panel, then set the strength.
Is my image uploaded?
No. Text recognition, the privacy model, face detection and code scanning all run inside this page. Its security policy only lets it talk to pixelate.run itself, and once the models are cached it works with the network switched off. If you choose fast cloud recognition instead of the on-device model, the text read from the image, and only that, is sent to pixelate.run, checked by an AI model on Cloudflare and not kept.
Can a pixelated number be recovered?
Not from an image made here. An ordinary mosaic or blur can sometimes be reversed by matching it against known fonts. Before text is pixelated or blurred, this tool shuffles the blocks at random, which leaves nothing to match. A solid block removes the pixels outright.
What about the location stored in my photo?
The saved image is drawn from scratch, so GPS position, camera model and every other piece of metadata stay behind.
Will it work on my phone?
Yes, in current versions of Safari, Chrome, Firefox and Edge. The first image takes a few seconds longer while the recognition models download (about 11 MB); after that they load from your device.
Why does it download a large model on a computer?
To tell a name, a nickname or an address from ordinary words, it uses OpenAI Privacy Filter, an open model of about 917 MB that runs in the page like everything else. It downloads once, in the background while the patterns give a first answer, and then stays on your device. Phones and tablets use the patterns alone.

More ways in