How to Unblur an Image

Every guide to unblurring photos opens by promising that AI can fix any blurry picture. That is not true, and believing it is why people upload a photo, get a disappointing result, and conclude the tools do not work.

The honest version is more useful: blur comes in several kinds, and they are not equally recoverable. Knowing which kind you have tells you what to expect before you spend a credit.

The three kinds of blur

Motion blur — usually recoverable

Your hand moved, or the subject did, while the shutter was open. The detail is still in the file; it has been *smeared along a direction*. Because that smearing follows a consistent path, a model can reason backwards along it.

This is the most recoverable kind of blur there is. A photo taken at 1/15s indoors, where everything has a slight directional drag, often comes back looking like it was taken at 1/125s.

Out-of-focus blur — partly recoverable

The lens focused on the wrong plane. Slight misfocus — the eyes soft but the ears sharp on a portrait — recovers reasonably, because there is still edge information nearby to work from.

Severe misfocus does not. If the subject is a smooth gradient with no discernible edges, there is nothing to reverse. Anything that appeared would be invented, and on a face that matters.

Compression and resolution blur — recoverable, but it is a different tool

This is not really blur. The photo has been saved at low quality or at a small size and then enlarged, so what looks like softness is actually missing pixels. Sharpening will not help — you cannot increase the contrast of detail that is not there.

The fix is upscaling, which reconstructs the missing pixels rather than trying to sharpen the ones you have.

How to tell which one you have

Zoom to 100% and look at a high-contrast edge — a dark branch against sky, the frame of a window, text on a sign.

• Streaks in one consistent direction → motion blur.
• Soft in every direction, evenly → out of focus.
• Blocky squares, or colour bleeding in rectangles → compression. Upscale, do not sharpen.
• Visible pixel steps → resolution. Upscale.

The order that matters

Most bad results come from doing the right things in the wrong order.

1. Start from the largest original you have. A photo that has been through three chat apps has less to recover than the one still on the camera roll. This is the single biggest factor and it costs nothing.
2. Denoise before sharpening. Sharpening raises local contrast, and noise *is* local contrast. Sharpen a noisy photo and you get grit that looks worse than the blur did. The photo enhancer handles noise and sharpness together, which is why it usually beats sharpening alone.
3. Sharpen once. Two passes stack halos on every edge and the result looks brittle rather than sharp.
4. Upscale last. Enlarging a soft image enlarges the softness with it.

Which tool for which job

The specialist tool beats the general one when its specific problem is the only problem. The enhancer beats everything when there are several problems at once, because the corrections inform each other.

What sharpening actually does — and its limit

Sharpening increases contrast at edges. A dark pixel next to a light pixel becomes darker and lighter. Your eye reads increased local contrast as "sharper", which is why it works at all.

The limit follows directly: it can only emphasise edges that exist. If an edge has been blurred until neighbouring pixels are the same value, there is no edge to emphasise. Turning the slider further just amplifies the halos around edges that survived, which is what makes over-sharpened photos look crunchy and artificial.

AI models go further because they have seen millions of sharp photographs and can reconstruct plausible detail rather than only emphasising existing detail. That is genuinely better — and it is also why you should look at the result carefully. On texture (fabric, foliage, hair) the reconstruction is close enough to be indistinguishable and genuinely useful. On text and faces it is a statistical guess, and a confidently wrong character is worse than a blurred one.

Five things that will not work

• Sharpening the same photo repeatedly. Halos compound. One pass, from the best source you have.
• Screenshotting instead of downloading. A screenshot is a re-capture at screen resolution, which throws away everything you were trying to recover.
• Enlarging first, then sharpening. The enlargement bakes in the softness.
• Expecting a face to come back from nothing. A face that a person cannot identify in the original is a face the model will reconstruct rather than recover.
• Working from a social media download. Every platform recompresses. Find the original.

The realistic expectation

A slightly soft photo becomes a good photo. A moderately blurred photo becomes usable. A badly blurred photo becomes a slightly less blurred photo, and no tool changes that — the information is not in the file.

That is not a limitation of any particular product. It is what blur is.

Try it on your own photo — the photo enhancer is free for the first three images, no card required, and takes about fifteen seconds.