Blurred or pixelated text can often be recovered, because the hidden words use a known font and only a few dozen possible characters; public tools such as Depix and Unredacter do exactly that. Use Blur only for text that is not secret, and cover anything private with the solid black block, which leaves nothing to reconstruct.
Blur hides text from people, not from software
When you blur a word, each pixel becomes an average of the pixels around it. Your eyes can no longer pick out letters. But the blur is a predictable, mechanical process, and the text underneath came from a small set of possibilities: letters, digits and punctuation, in one of a handful of common fonts, at a size you can measure from the surrounding text.
That combination is what reconstruction tools exploit. They do not "unblur" anything. They guess: render a candidate letter in the same font and size, blur or pixelate it the same way, and compare it with the hidden area. Keep the candidate that matches best, move to the next letter, repeat.
- Depix (a 2020 proof of concept) rebuilds words hidden by box-filter pixelation in screenshots, by rendering candidate text in the same font and matching blocks. Its README is open about the limits: you need to know the font settings, and recompressing the image afterwards breaks the method.
- Unredacter (Bishop Fox, 2022) took the same idea further, guessing one character at a time and checking each guess against the pixelated region, and recovered the text in a pixelated image that the security firm Jumpsec had posted as a public challenge. The researchers' conclusion: when you redact text, use black bars over the whole text, and nothing else.
Text is the easiest target because the alphabet is small and the font is usually visible elsewhere in the same image. That is why this page opens with Solid block selected and labels Blur and Pixelate as decorative.
When blurring text is fine
Blur has honest uses where nothing is secret:
- toning down busy background signage so a product or person stands out;
- softening the rest of a page in a tutorial screenshot so readers look at one button;
- hiding a spoiler or a punchline in a meme where the "secret" is not sensitive;
- making a brand name less prominent in a casual photo.
If you would be uncomfortable seeing the words read back, use the solid block.
Hide text in a picture
- Load the picture through the file button or by dragging it in.
- Click Find text to have the page read the image on your device. It boxes the kinds of text most often private: anything shaped like an email address, phone or card number, IBAN or long account number, plus QR-type codes.
- Drag boxes over the other words you want hidden, such as names, street names, usernames and messages. Make each box cover the entire word or line, with a little room around it.
- Choose how to cover them under Cover with, one setting for every box in the picture: Solid block if any of the text is private, Blur only when all of it is decoration.
- Press Export and save the new copy. Zoom in once to make sure no letter edges peek out.
Small habits that make hidden text stay hidden
- Cover whole words and lines. Hiding the middle of an email address or the second half of a name often leaves enough to guess the rest.
- Do not let the box size tell the story. A box that fits a name exactly reveals its length, and sometimes the shape of tall and short letters at its edges. Use uniform, slightly oversized boxes when hiding several names.
- Hide every copy. The same name may appear in a header, a signature, a reply quote and a notification banner.
- Check text inside other images. Profile pictures, embedded photos and logos may contain text too.
- Avoid translucent markers. Highlighter-style strokes in photo apps are partly see-through by design; raising the contrast can bring the letters back.
What Find text can and cannot find
Find text runs an OCR engine (Tesseract) in your browser and then applies pattern rules. It is good at clean, high-contrast screen text and printed documents. It looks only for specific patterns, because a rule cannot tell which ordinary words are private. Names, addresses written in words, messages and passwords shown in plain text all need a box drawn by hand.
OCR also struggles with handwriting, curved or angled text, very small or light-gray print, text over photos, and decorative fonts. Use it as a first pass, never as the only check.
The result you download
Export repaints the image with your boxes and writes a new file:
- covered areas hold either a black block or the decorative blur, nothing more;
- no metadata from the original travels with it, so camera details, dates, GPS coordinates, captions and editing history are not in the new file;
- the format stays JPEG, PNG or WebP as you opened it (HEIC, opened in Safari, becomes JPEG);
- the image keeps all of its pixels; only if a phone cannot encode something that big does the page fall back to a 4,096-pixel version and say so.
Supported images and browsers
- JPEG, PNG and WebP in current Chrome, Edge, Firefox and Safari. HEIC in Safari only.
- Phones and tablets work; you draw boxes with a finger.
- You can load several images and step through them.
- Scanned PDFs, animated images and video are not supported.
The image does not leave your browser
Everything on this page happens in the open tab. The image is read from your device, the OCR engine runs locally, and the finished copy sits in memory until you save it. The OCR engine and English data arrive from this site on your first use of Find text and never include your picture. Closing the tab clears the page's copies.
How it works
Boxes are stored in the original image's pixels and painted onto a fresh canvas at export, then turned into a new file by a worker thread when the browser has one. The blur option shrinks each box's content to a few pixels and scales it back up twice, which looks soft but is still an average of the real pixels, hence the "decorative" label. To watch for yourself that no request carries your image, open your browser's developer tools as described in the step-by-step check.
FAQ
Can blurred text be read?
Often, yes. Reconstruction tools guess characters in the same font, blur or pixelate each guess, and compare. They work best on screenshots, where fonts are standard. A solid block cannot be read because nothing of the text remains.
Is a heavier blur safer?
A stronger blur raises the effort but does not remove the weakness, since the result still depends on the original letters. Use the solid block for anything private.
Does Find text detect all text?
No. It reads all the text but only boxes patterns that are usually private, such as email addresses and long numbers, plus 2-D codes. Box other words by hand.
What color is the solid block?
Black. It is fully opaque, and the pixels underneath are replaced when you export.
Will the person I send it to see that something was hidden?
Yes. A black box makes it obvious that something was removed. That is usually a good thing: nobody mistakes it for the original.
Does my image get uploaded for OCR?
No. The OCR program comes to your browser, not the other way round; your picture is never sent.
Sources
Checked 2026-10-11.
- Depix README on Codeberg: how box-filter pixelation is matched, and the stated requirements.
- Bishop Fox blog post introducing Unredacter: the Jumpsec challenge and the solid-bar recommendation (February 2022).
Last updated: