BlurThis

Guide

How to unblur an image (and when it is simply gone)

By Shaibaz, graphic designer and creator of BlurThis

Unblur apps sharpen and guess. A solid bar has nothing left to restore. Generative unblur is an illustration, not recovery.

Laptop showing a blurry photo beside a sharper version, magnifier on the desk
Illustration by BlurThis

BlurThis does not unblur images. This page is the recovery question people type after they soften a photo, after they find an “AI unblur” site, or after someone claims a redaction can be undone.

Improving a soft photograph (missed focus, heavy shrink) is not the same as estimating missing detail with a generative model, and neither is the same as recovering text that someone deliberately covered. Soft photos can sometimes look sharper. Generative tools often invent plausible pixels. A solid black bar that replaced the original pixels has nothing left to restore. If your job is covering a secret before you share, start from how to blur a screenshot and blur vs black bar.

What do unblur tools actually do?

Almost all of them do one of three things. They sharpen, which raises local contrast so edges read crisper. They upscale, which fills a bigger canvas with a model’s best guess at detail. Or they try to reverse a mild, fairly even blur. That last case is the only one that is genuinely undoing anything, and it needs the blur to be light.

A gentle motion smear on a licence plate is the best case for recovery research. A heavy smear over a phone number usually is not, because the information has been averaged away rather than hiding faintly underneath. Short strings are the risky middle ground: enough structure may remain to narrow candidates even when a human eye says “blurry.”

Why a generative “AI unblur” looks so convincing

Because it is often drawing a new image that fits the blur it sees, not reading the original secret back. Feed the same blurred face to two tools and you can get two different people, both sharp, both confident. That is the tell.

A face in a photo with a box drawn around itOpen larger image in a new tab

It matters if you are trying to verify something. A generated digit is not evidence the original digit survived. Treat the output as an illustration, not a recovery. If you would act on it, you need the original file, not a model’s guess.

When can blurred text really be read back?

When the text is short, the alphabet is small, and the blur is light. Researchers demonstrated this in a 2016 PETS paper on mosaicing and blurring as document redaction: the structure left behind is often enough to narrow the candidates dramatically. Six digits from a set of ten is a small search space. That is leftover information, not magic.

A paragraph of body copy under a thick smear is the opposite case, and usually stays gone. So the risk is not evenly spread. It concentrates exactly where you care most: codes, PINs, card numbers, short surnames. We wrote that up in the PETS 2016 result.

A highlighter tint is a third failure, and it is not even blur. 9to5Mac showed Markup strokes lifting with Photos brightness. Do not call that “unblur research.” Call it leftover contrast. Markup brightness reveal covers that path.

What a solid black bar means for recovery

If the cover fully replaced the pixels and you exported a flat image, there is no original text left in that region to restore. Unblur tools cannot pull digits out of a filled rectangle. What they can do is invent something that looks like digits. That is not recovery.

Side by side soft smear and solid black bar over fake textOpen larger image in a new tab

JPEG ringing around a light smear can look like a recovery. It is not. Start from a flattened opaque fill when the secret is a code. Details in JPEG vs PNG for redacted screenshots.

Should you upload an image to an unblur site?

Think about what you are handing over. The picture you want unblurred often includes more than the soft patch: a chat header, a timestamp, an account name in the corner.

For a holiday snap that came out soft, some people still choose an online tool. For a document, an ID, a bank screen, or anything from work, treat the upload as sharing that file. Read the service’s processing and retention policy before you send it. Do not assume a copy is deleted just because a page said so. Prefer editors that keep the file in the tab when the content is sensitive: blur online without upload.

How to make a cover nobody can undo

  1. Stop relying on blur for the sensitive bits. Name the secret: OTP, card, cookie, phone, recovery code.
  2. Paint a filled black bar that replaces the pixels. Pad past the edges. What you should see at 100% zoom is a uniform rectangle, not a faint stem.
  3. Export a flat PNG. Open the download. Raise brightness and contrast. If a shape reappears, that was a tint, not a redaction.
  4. Attach the export. Keep the original off the thread. Rotate anything that was ever on a file that left the building.

Use blur where the goal is “not identifiable at a glance”: a bystander’s face, a background whiteboard. Use a bar for anything you would change or cancel if it leaked. For a browser editor that keeps the file on your device, start on the BlurThis homepage.

When “unblur” is a sales pitch, not a recovery

Marketing pages for unblur tools often show a soft face that snaps back into a crisp portrait. That demo is usually generative fill: the model invents eyes and skin that look plausible. It is not proving that your black rectangle still encodes the original digits. If the cover replaced every sample in the region, there is no latent face left to reconstruct. The persuasive before-and-after is a different file path.

Researchers who study mosaicing and light blur on text (the PETS 2016 line of work) care about residual structure, not about recreating a celebrity from a solid fill. Treat those papers as a warning against polite smears on OTPs, not as a promise that every cover can be undone. For the research walkthrough, read PETS mosaicing and blur research. For the iPhone tint failure that fooled people in 2018, see Markup brightness reveal.

If your goal is to keep a secret out of a thread, stop shopping for unblur. Paint an opaque bar, export PNG, and attach only that file. If your goal is to rescue a soft holiday photo, say so honestly: you want enhancement, not redaction forensics. Those are different products and different risks when the pixels include other people's data.

Recap

Sharpening and generative “unblur” are not magic recoveries of deliberate covers. Light blur on short text can leave structure. Highlighter tints can lift. A solid bar that replaced the pixels has nothing left to restore. Paint the cover once, export PNG, and do not ship the soft original.

Related guides

Back to all guides