Three blurry photos, three different accidents

Three photos sit side by side on your phone. The first is Thanksgiving dinner, shot handheld under warm dining-room lights. The second is your dog halfway across the dog park. The third is your kid at bat in a Little League game, taken as the light was going. At phone size, all three fit one word: blurry. There's no reason to tell them apart, until you try to save them.

Zoom in, and they tell three different stories.

In the Thanksgiving photo, everything in the frame slid the same way. The lettering on the wine label smears in one direction. The edge of the table smears the same way, and a candle flame has been pulled into a short line leaning the same way too. The whole frame moved at once.

In the dog park photo, only the dog is dragged. The grass, the fence and the people on the bench are as sharp as ever. The smear stays inside the part that moved and stops right at its edge.

In the Little League photo, look for a smear and you won't find one. The chain-link fence in front of the batter is crisp, and the batter is soft, evenly soft, with no direction at all, like looking through slightly fogged glass. The camera focused on the fence.

These aren't three strengths of the same problem. Each one leaves a shape of its own, and that shape, more than how bad the photo looks, decides how much can be brought back.

Read the blur in 30 seconds: which way it runs, and where

You need two questions, and you can answer both by eye.

1. Which way does it run? Zoom in on something with hard edges: text, a door frame, the rim of a glass, a lamp. Is there a smear, a doubled edge or a streak that runs in one direction? Or is everything soft without any direction?
2. Where is it? Across the whole frame, or only on one thing?

Put the two answers together:

• The same streak everywhere, all leaning the same way: camera shake.
• Streaks on one subject while the background stays sharp: subject motion.
• No streaks, just softness, often with something else in the frame crisp: missed focus.
• Edges still crisp, but the photo is speckled and the dark areas look blotchy: not blur at all, but grain from shooting in low light.

The rest of this article explains why those four answers lead to four very different results.

Four kinds of "blurry" and what causes each

Camera shake: the whole frame slides one way

Researchers from MIT and the University of Toronto, presenting Removing camera shake from a single image at SIGGRAPH 2006, state the problem plainly: "Camera shake during exposure leads to objectionable image blur and ruins many photographs." Their method treats shake as one blur shared by the whole picture: it "assumes a uniform camera blur over the image and negligible in-plane camera rotation."

That assumption is the useful part for you. When the camera itself moves, the whole scene moves across the sensor together, which is why every edge in the Thanksgiving photo smears in much the same way.

Subject motion: one thing smeared, everything else sharp

Here the camera held still and the subject didn't. The smear has a direction, but only inside the part that moved. That breaks the single, frame-wide blur described above: the photo now holds two different rules, a moving dog and a still park, in the same frame. That's why this is the hardest case to repair.

Missed focus: soft everywhere, no direction

A depth-of-field lesson on Stanford's CS 178 course pages, written by Marc Levoy, defines "out of focus" precisely: "The standard definition is that the width of the blurred image of the feature has become larger on the sensor than some maximum allowable circle of confusion." Points that come to a focus in the wrong place "will have spread out to a size larger than the circle when they strike the sensor."

The key word is circle. An out-of-focus point spreads evenly in every direction instead of being dragged along a line, so there's no streak to see and no direction to follow.

Not blur at all: grain from low light

The school play in a dark auditorium can look blurry on a phone screen. Zoom in, though, and the edges are there. What's wrong is a speckled layer of grain over everything, worst in the shadows. That's noise, not blur, and it needs a different fix. The full walkthrough is in our guide to fixing grainy night photos.

Why the direction decides how much comes back

A standard way to describe a motion blur uses just two numbers. MathWorks' documentation for MATLAB's fspecial function includes a "motion" filter that "Approximates the linear motion of a camera" and takes a length, measured in pixels, and an angle, which is "measured in a counter-clockwise direction from horizontal". Its deblurring tutorial builds a blur "specifying linear motion across 21 pixels at an angle of 11 degrees" and then restores the image using that same description.

That's the whole idea in miniature. A smear with one direction and one length can be described, and what can be described can be worked backward. A missed-focus blur offers no direction to describe, because every point spread into a circle, so there's much less to work back from. Subject motion sits in between: it has a direction, but the direction only holds for part of the frame.

The same tutorial also shows why grain changes everything. Without an estimate of the noise, it says, the restoration filter "is equivalent to an ideal inverse filter, which can be extremely sensitive to noise in the input image." In its own example, "The noise in this restoration is amplified to such a degree that the image content is lost." A deblurring step can make grain worse rather than better, which is why a grainy photo needs denoising instead.

The tool you use may not ask you for an angle or a length. It can still only work with what's left in the photo, and a smear with one consistent direction leaves more behind than a point that spread out evenly in every direction.

What to expect from Deblur and denoise

Deblur and denoise has exactly two modes, and you pick one per run: Deblur, labeled for "Blurry, camera-shake photos", and Denoise, for "Grainy, low-light photos". Apart from that choice, there's nothing to set, and the tool works on the whole photo; there's no way to pick out one area. So the decision that matters most is the one you make in the first column below.

What you see	Most likely cause	What to realistically expect	Mode to pick
The whole frame slides one way; text and edges smear in the same direction	The camera moved when you pressed the shutter	Mild to moderate shake: much of it usually comes back. Strong shake: only partly better, and rarely crisp	Deblur
One subject is streaked while the background stays sharp	A person, a pet or a car was moving	The hardest case: the result usually looks cleaner but loses small detail in the part that moved	Deblur
Soft everywhere, or soft on the subject with no streaks	Focus landed somewhere else	Some clarity comes back; areas that were fully out of focus don't regain real sharpness	Deblur
Edges look fine, but the photo is speckled and dark areas are blotchy	Low light forced the camera to boost its sensitivity	This is Denoise's job; Deblur barely changes it	Denoise

Those four rows aren't four facts to memorize. They follow from the same rule as the section above: the clearer the direction, the more comes back.

One honest sentence applies to every row. When a photo is so blurred that you can't find a single sharp edge anywhere in it, the improvement is real, but nobody can promise it will look like a photo that was in focus from the start, and this article won't promise that either.

If you also want the photo bigger, that's a separate job for Upscale image. When a run uses credits, the editor tells you the cost before you start it. If you're still not sure which tool fits your photo, this guide to picking a tool for a blurry photo sorts the options side by side.

Four places to check the result

Don't judge the whole picture. The whole picture nearly always looks better than the original; that's what these tools do best, and it's also where you're easiest to fool. Look at four places instead:

• Straight lines and text. Signs, door frames, table edges, the spines of books. People notice mistakes on straight lines faster than anywhere else, so problems show up here first.
• The edge between the subject and the background. A subject that got sharper in front of a background that didn't can look cut out and pasted on, as if two photos were stacked together.
• Small identifying details. The pattern on a shirt, the lettering on a box, the design on a tile. This is where a tool may fill in something plausible but wrong: it reads as text at a glance, and up close the letters aren't the ones that were there.
• The dark areas. Damage that was hiding in the shadows from the start often shows up only after the rest of the photo is cleared up.

The result screen keeps your original and the result together, layered with a draggable divider on a phone and laid out as two panels on a bigger display. Either way, work through those four places in order rather than glancing over the whole frame.

Glance at the corners too: whether a downloaded file carries a little vipix.vn mark depends on the tool and on whether you were signed in when you ran it.

Kids and pets: a cleaner photo that looks slightly off

These are the photos people most want to save, and they land in the hardest row of the table.

It happens the same way in every home. Your kid is laughing before you've got the phone up. The cat turns its head right as you tap. Indoor light is never quite enough. Together that produces row two, a smeared subject against a sharp background, with one more problem on top: the face often takes up only a small part of the frame, so there's little information to work with.

Here's what often happens, so it doesn't catch you off guard: the result looks much cleaner, but the expression looks slightly different. Not distorted, not someone else, just slightly different. The reason isn't mysterious. The detail the blur destroyed is gone from the photo, and what fills the gap is a best guess built from what's left. For the edge of a table, that guess is usually close enough. For a face, you aren't comparing the result with the original file; you're comparing it with your memory of that person, and memory is a much stricter judge.

So the advice is simple: keep the original next to the new version, and don't replace the old file with the new one. A photo like this usually can't be taken again, and two versions to compare beat one nice-looking version with no way back. If someone in the family looks at the new one and says right away that it looks off, trust them over yourself. You've stared at it dozens of times; they're looking with their memory.

Two situations call for a different tool. When everything around a face is crisp and the face alone has gone soft, switch to Enhance faces, which works on faces specifically and tries to keep the features looking natural. And if a face looks broken because the photo passed through several phones before it reached you, the cause is something else entirely: why photos you get by text look blurry.

One more thing worth a thought when you're in a hurry. Saving a damaged photo often means uploading it to several places in one evening, and each upload leaves another copy on someone else's server. Vipix keeps uploads and results only for a limited time; the periods are set out in Vipix's Privacy Policy.

When the blur is the point

Not every streak is an accident.

If you follow a running subject with the camera, you get the reverse of row two: the subject stays sharp and the background streaks. By the table, that photo counts as blurry, but here the blur is the point. It's how a still picture shows movement. Run it through Deblur and you'd be cleaning away the very thing that makes it work, and you'd get back a tidy, frozen picture with nothing left to look at. The same goes for Fourth of July fireworks: the trails are the picture.

On the other hand, a smear right across a face, or across the one product the photo was taken to show, is real damage. The same amount of streaking can mean opposite things depending on where it sits.

Between those two ends there's a zone you can save yourself the trouble over. A smear at the edge of the frame, on the wall to one side or on the sleeve of someone standing at the end of a row: leave it. Nobody looks there, and because Deblur and denoise works on the whole photo rather than one corner, changing every pixel to fix a spot nobody looks at isn't a good trade.

If the photo might become evidence, keep the original

There's one more reason to keep the untouched file. In State of Washington v. Puloka, according to a May 2024 alert from the law firm Greenberg Traurig, a Washington state superior court judge rejected an AI-enhanced version of a bystander's smartphone video of a 2021 shooting. The defense's expert had argued that the source video "was low resolution, had substantial motion blur", and the enhanced version was made with an AI video tool from Topaz Labs. The state's expert testified that the enhanced video "added information that was not in the original files", "removed artifacts on individual images" and "altered shapes and colors in the video". The court found that it "did not show with integrity what actually happened, but instead used opaque methods to represent what the AI model thought should be shown".

Read it at its actual size: one trial court in one state, ruling on video offered as evidence in a criminal case. It isn't a rule about photos in general, and this isn't legal advice. The practical lesson is narrow. If a photo might matter later (a fender bender, a package that arrived damaged, a dispute with a landlord), keep the original file untouched and share that one when it counts. An enhanced copy is fine for looking at; the original is the record.

Next time: stop the blur before it happens

Fixing blur afterward is always a compromise, so a few habits are worth more than any tool. None of them needs a setting or a number.

• Against camera shake: hold the phone with both hands and keep your elbows against your body or lean on something solid. Press the shutter gently. For a still subject, the self-timer takes your finger out of the picture.
• Against subject motion: wait for the moment the subject pauses, or take several shots in a row and keep the sharpest.
• Against missed focus: tap the subject on your screen before you shoot, so the camera knows what matters.
• Against grain: add light where you can. Turn on a lamp, open the curtains or move closer to a window.

Frequently asked questions

How can I tell camera shake from an out-of-focus photo?

Zoom in on a hard edge such as text or a door frame. Camera shake leaves a smear or a doubled edge that runs in the same direction across the whole frame. Missed focus leaves softness with no direction at all, often while something else in the frame is sharp.

Can AI fix motion blur?

Partly. Mild to moderate camera shake comes back the most, and a moving subject is the hardest case. A heavily blurred photo does improve, but no tool can promise it will look like a shot that was sharp in the first place.

Can an out-of-focus photo be fixed?

Up to a point. Some clarity returns, but areas that were fully out of focus don't regain real sharpness, because the detail spread out evenly in every direction instead of along one line.

Why does my child's face look a little different after deblurring?

The detail the blur destroyed has to be filled in with a best guess based on what's left, and you're comparing the result with your memory of that face. Keep the original next to the new version. If the rest of the scene is sharp and only the face is soft, a tool made for faces, such as Enhance faces, is the better fit.

Deblur or Denoise: which one should I pick?

Streaks or smears call for Deblur. Speckled grain calls for Denoise. Each run uses one mode, so a photo that's both smeared and grainy takes two runs.

Start with the photo you'd hate to lose

Pick the blurry photo you'd hate to delete, read which way its blur runs, then try Deblur on it, or Denoise if it turned out to be grain.