Can blurred photos be unblurred?
Sometimes, yes. Whether a blurred face or number can be recovered depends on how it was hidden, how strongly, and whether the original is still inside the file. Here’s what has been reversed in practice, and how to blur so it can’t be.
By the BlurTap team · Updated
Short answer: a light blur, fine pixelation or a swirl effect can often be reversed or matched to the original, especially on text. A heavy blur, coarse pixelation or a solid box, flattened into the image, can’t be undone: the information is gone. AI “unblurring” of a heavily blurred face produces a made-up face, not the real one.
How each method holds up
| Method | Can it be reversed? | Why |
|---|---|---|
| Swirl, twirl or warp | Yes | It moves pixels around without destroying them, so it can be run backwards. |
| Fine pixelation (many small blocks) | Often, for text and faces | Each block keeps an average of the detail. Text can be matched letter by letter. |
| Light blur | Partly | Edges and shapes survive. Software can still recognise who it is. |
| Semi-transparent highlighter | Yes | Raising contrast or exposure brings the detail back. |
| Black box in a PDF or design file | Yes, if not flattened | The box is a layer on top. The text underneath is still there. |
| Heavy blur, flattened | No | Detail smaller than the blur is destroyed. |
| Coarse pixelation (few large blocks) | No | Too little information is left to match. |
| Solid opaque cover, flattened | No | The original pixels are replaced. |
Cases where hiding failed
The swirled face (2007)
Interpol released photos of a suspect who had hidden his own face with a swirl filter, after German police specialists reversed the effect. The public appeal led to the arrest of Christopher Paul Neil in Thailand. Swirl is a distortion, not a deletion, and that is why it could be undone.
Recognising pixelated and blurred faces (2016)
Researchers from Cornell Tech and the University of Texas at Austin trained standard neural networks to identify people in pixelated and blurred images, including faces blurred with YouTube’s own blur tool. Their paper, “Defeating Image Obfuscation with Deep Learning”, found recognition rates far above chance for the obfuscation levels then in common use. The network didn’t need to reconstruct the face. It only needed to match it.
Pixelated text in screenshots (2020)
Depix, an open-source proof of concept, recovers text that was pixelated in screenshots, such as passwords and account numbers. It works because screenshot text uses a known font at a known size: the tool pixelates candidate characters the same way and compares the blocks.
Black boxes that weren’t
Court filings and government documents have repeatedly been published with black redaction boxes drawn on top of text that was still in the file. Copying and pasting the “redacted” passage revealed it. The same applies to photos edited in apps that keep edits as a separate layer.
How to blur so it can’t be undone
- Go strong. A blurred face should be an indistinct smudge, not a soft-focus face. With pixelation, aim for no more than about 6 to 8 blocks across a face.
- Use a solid cover for text: numbers, names, addresses, barcodes and QR codes. Or crop them out.
- Never use swirl or warp effects to hide anything.
- Flatten before sharing. Export a new image. Don’t send design files, PDFs with drawn-on boxes or originals with edit history.
- Turn off Live on an edited Live Photo before sharing. The moving part is a separate short video.
- Hide the context too: tattoos, distinctive clothing, house numbers and the location data in the file. See how to remove location on iPhone.
On iPhone, Photos keeps your edits reversible in your own library (Revert), but the copy you share is a flattened image. The edit can’t be undone by whoever receives it. How safe it is still depends on the method: see every iPhone method compared.
Frequently asked questions
Can AI unblur a face?
AI “unblur” and “enhance” tools can turn a blurred face into a sharp one, but when the blur is strong the result is invented: the model draws a plausible face, not the real person. The danger is with weak blurs and fine pixelation, where enough detail survives for software to recognise or match the person.
Is pixelating safer than blurring?
Neither is safe by default. Both depend on strength. Coarse pixelation with a few large blocks across a face, or a heavy blur, destroys the detail. Fine pixelation and light blurs leave enough of it to be matched, especially on text, where the set of possible letters is small.
Can someone remove a blur from a photo I posted on Instagram?
If the blur was applied to the image itself before posting, there is no separate layer to remove. All that’s left is the blurred pixels. Whether they reveal anything depends on how strong the blur was. Edits stored as a separate layer, as in some PDFs and design files, are a different story: the original can simply be uncovered.
What’s the safest way to hide text in a screenshot?
Crop it out if you can. Otherwise cover it with a solid, fully opaque box and share a flattened image (a screenshot of the edited screenshot works). Avoid pixelating text: tools exist specifically to reverse pixelated text in screenshots.
Does BlurTap’s blur hold up?
BlurTap applies a strong blur and writes it directly into the pixels of a new image, so the shared copy contains no trace of the original detail and no layer to peel off. Your original stays untouched in your library.