Which kind of blurry is your photo?

Three photos, one word. The kids in costumes on the front porch at dusk came out soft and speckled. The shot you zoomed in on from the top of the bleachers at the Little League game looks blocky. The picture your cousin texted from the family reunion is sharp everywhere except the faces. All three get the same label — blurry — and all three tend to go into the same tool, which is how people decide that AI photo fixing doesn't work. More often, two of the three went through the wrong door.

"Blurry" covers at least four faults that start in different places: a shortage of pixels, a smear from shake or missed focus, a layer of grain and faces that went soft while the rest of the frame stayed sharp. Name the fault first and the right tool follows — and sometimes the honest answer is a new photo rather than a fix.

Two things to try before you open any tool

If you took the photo, check the lens. A picture that looks hazy all over, as if it were shot through a fogged window, often comes down to a smudge on the lens rather than anything a tool should fix. Phone makers such as Motorola treat it as step one: on its North American help page for camera issues with the moto g play, the first line under the heading Photo looks blurry reads, "Wipe the lens clean with a soft, dry cloth and retake the photo." If new photos still look soft, the same page points to a setting: "Increase resolution in the Camera app to the highest available. A lower resolution results in lower image quality."

If someone else took it, ask for the original. School portraits, team photos and event galleries often reach you as a download that was squeezed to load faster. Snapfish, the photo-printing service, describes the giveaway in its help center: "If your photo's resolution is high, but the file size is small, it is an indication that the file is compressed, and any prints or products created with it will appear blurry. This is frequently a problem with photos obtained from professional photographers, and the best solution is to request uncompressed versions of the photos." A photo that came to you in a text thread may have been shrunk on the way too — why texted photos look blurry explains what happens to the file and how to ask for a better copy.

Both take a minute, and both beat any repair: a retake or an original file holds detail that really exists, while every tool below can only work with the copy you have.

Zoom to 100% and name what you see

Diagnosis doesn't need jargon, just the right magnification. Open the photo at 100% — the view many photo apps call actual size, where one pixel of the photo fills one pixel of your screen — and look at a spot with fine detail: lettering on a sign, a few strands of hair, the edge of a leaf.

The US print industry already has words for what you'll find. Blurb, which prints photo books, lists four warning signs in its help center: "Blurriness or softness — nothing is clearly in focus. Grain or noise — a speckled or rough texture, often caused by a high ISO camera setting. Pixelation — visible square blocks where detail should be. Jagged edges — stair-step artifacts on diagonal lines or curved edges." Those signs, plus one case that only affects faces, make up the table below.

What you see at 100%	Usual cause	Name of the fault
Detail smeared in one direction; points of light stretched into short dashes	the camera moved during the shot, or the subject did	motion blur
Soft all over, with no direction to the softness	the camera focused on the wrong thing	missed focus
Fine speckles, worst in the dark areas	low light, so the camera raised its sensitivity	noise (grain)
Fine at small size, square blocks and stair-step edges once enlarged	too few pixels: digital zoom, a small file or a compressed copy	low resolution
The scene is sharp, but the faces are soft	faces small in the frame, far away or copied many times	soft faces

The first two rows end up at the same tool, but they're worth telling apart because they point to different habits the next time you shoot. A quick tell: motion blur has a direction — find a point of light and you'll see it stretched into a streak — while missed focus is evenly soft everywhere. When you need to know which one you have and what caused it, telling camera shake from a moving subject is its own diagnosis.

The fourth row has a cause phone makers spell out. On the same Motorola page: "Avoid zooming when taking a photo. Pixelation always occurs when you zoom a digital photo. If you use zoom during the shot itself, pixelation will be more evident." That's the shot from the back row of the school play, or from the top of the bleachers. Next time, walk closer instead of zooming in.

A photo that's hazy from corner to corner isn't in the table: that's usually the smudged lens from the first section, and no tool below is the fix.

Going to print? Check again at 200%. Blurb asks its customers to zoom to 200% before they order, and its reasoning is blunt: "If it looks poor at 200% it will look that way (or worse) in print." For an image that fails, its advice doesn't soften either: "If an image fails the 200% test, replace it with a sharper original. There's no reliable way to repair a poor-quality image, sharpening filters in Photoshop or similar tools can help marginally but will make things worse if overdone." Keep that line in mind for the rest of this article; it marks the honest limit of every tool in the next table. How many pixels a print needs depends on its size and on how far away people will stand, and how big you can print a photo works through the numbers.

Match the fault to the tool

Once the fault has a name, picking a tool is a lookup. The last column matters as much as the third: most disappointment comes from using a tool on a fault it wasn't built for, not from the quality of what it does.

Fault	Tool	What it does	What it won't do
Low resolution — the photo is clean, just small	Upscale image	At 2K, 4K or Extreme Scale, enlarges the photo toward that level's output size, building plausible detail and sharpening edges	Rescue shake or grain; it's built for photos that are already fairly clear
Motion blur, missed focus or grain	Deblur and denoise	Clears smearing (Deblur mode) or speckled grain (Denoise mode) and keeps the photo at its original size	Enlarge the photo; very heavy blur improves only partly
Only the faces are soft	Enhance faces	Works on every face it detects in the photo, in one run, with no settings to choose	Stand in for Deblur on a photo that's soft all over; for a crowded group shot, see how to sharpen faces in a group photo
Scratches, creases, yellowed or faded color — damage, not blur	Restore old photos	Removes scratches, rebalances faded color and sharpens, all in one run	Stand in for Upscale image on a photo that's in good shape and only needs to be bigger

Two details the table leaves out. Deblur and denoise holds two separate modes, and each run uses one of them; there's no setting that does both at once. Upscale image has four levels — Economy, 2K, 4K and Extreme Scale — that differ in the size of the file you get back and in how thorough the pass is, and none of them stretches the tool to a different kind of fault. When a run uses credits and the tool asks you to choose a level or a mode, every option shows its estimated cost per image before you run it.

The table also leaves out, on purpose, the idea of one tool that fixes everything. Enlarge a photo that camera shake has smeared and you get a bigger smeared photo.

Two faults in one photo: what to run first

Real photos rarely have just one problem. A phone snapshot of an old print can be small and soft, with a face that has lost its detail, all at once. Then the order you work in matters more than any single choice.

Dusk and indoor shots: smeared and grainy

Low light forces a trade, and Nikon's guide to low-light Halloween photos spells out both sides. On one side: "Sharp low light photography requires that the camera shutter stay open just a bit longer than when capturing in bright light, so keep a steady base of operation." On the other, if you'd rather keep the shutter quick, the same guide says to "Kick ISO higher (from 400 to 800) to work with a faster shutter speed." As Blurb's list above puts it, grain is often caused by a high ISO setting. That's why the photo of the kids in costumes on the front porch, taken as the light went, so often carries both faults: a little shake and a lot of speckle.

For a photo like that, run Deblur first, look at the result, then run Denoise on the result. One mode at a time shows you which step actually helped, and it lets you stop as soon as the photo is good enough. When grain is the main problem, cleaning up grain in night photos covers that case on its own.

Small and soft: clean it up, then enlarge

Enlarging is amplification. It makes good detail clearer and flaws more obvious. The Scientific Working Group on Digital Evidence (SWGDE), a US group of practitioners who publish best-practice guidance for handling images as evidence, puts it plainly in its image-processing guidelines: "Caution: A high degree of enhancement can also increase the visibility of existing noise and artifacts." So clean the photo first with Deblur or Denoise, and enlarge it with Upscale image in a separate run afterward.

Portraits and group photos: faces go last

Work on the whole photo first and save Enhance faces for the last run. Done the other way around, the whole-photo pass that comes later can run over the faces you just brought back.

Between runs, compare each result with the version you started from — on a phone you drag a divider across the picture, and on a wider screen the two sit side by side — and only then decide whether to run the next step. Fixing a photo is a series of small steps. Run three in a row before you look, and you won't know which one went wrong.

What sharpening adds, and why you keep the original

AI sharpening builds plausible detail out of what's left in the photo; it doesn't recover detail the camera never recorded, because no copy of that detail exists anywhere. What you get back is the most likely version of the scene, not a record of it.

There's also a ceiling on enlargement. SWGDE's guidance on resizing images notes that "when resizing by a factor of 10, information in a single pixel may be interpolated into 100 pixels." Push a tiny photo that far and most of what you see has been filled in. The same document warns its readers: "Practitioners should be cautious when utilizing novel interpolation methods, as it can be challenging to identify what processes were applied to the imagery and replicate those steps with accuracy." An AI-sharpened copy is a new rendering of the photo, not a measurement taken from it.

That distinction has already mattered in a US courtroom. A June 2024 post on the Duane Morris law firm's AI blog describes a ruling by a Washington state court judge: "In Washington v. Puloka, following a Frye hearing, the Judge excluded AI-enhanced video from being considered as evidence." The post adds that "it was important to the Judge's opinion that there was another version of the video (the original) that was available and usable – even if it was low resolution, with motion blur." It also notes that "A different result may prevail where there is no other alternative". That's one judge, one case and a video, not a rule for every court. The practical lesson still travels: when a photo serves as a record — an insurance claim, a dispute, a police report — hand over the original and treat the sharpened copy as the one you look at. This is general information, not legal advice.

The same working group's image-processing guidelines describe the habit that makes this easy: "The original image is preserved and processes are performed on a working copy." Keep the original file untouched, and run every fix on a copy.

Faces need one more check. Enhance faces is tuned to keep a person looking like themselves rather than smoothing them into someone else, but when a face is badly blurred, a mole, a scar or the fine lines around the eyes can come back slightly different. For a photo with a formal job — a memorial portrait, a print for a service — compare the face closely with the original before you download the result.

Frequently asked questions

Can AI fix a photo that's completely out of focus?

Partly. When nothing in the frame was ever in focus, Deblur can make the photo clearer, but don't expect it to come out as sharp as a well-focused shot. If the moment can happen again, a retake will do better than any fix.

Should I sharpen or upscale first?

Fix the blur or grain first, then enlarge in a separate run. Enlarging magnifies whatever is in the photo, flaws included, and SWGDE's image-processing guidelines warn that a high degree of enhancement can make existing noise and artifacts more visible.

Why does my zoomed-in phone photo look pixelated?

Digital zoom enlarges part of the frame rather than capturing more detail, so fewer real pixels cover your subject. Motorola's help page says it directly: "Pixelation always occurs when you zoom a digital photo." Next time, move closer instead of zooming in.

Will sharpening change how someone's face looks?

It can, a little. Enhance faces aims to keep a person recognizable rather than prettier, but in a badly blurred face, small marks such as a mole or a scar may come back slightly different. Check the face against the original before you download the result.

Is an AI-enhanced photo still the original?

No. It's a new rendering built from the original, with detail the AI filled in. Keep the original file, and if a photo ever has to serve as a record, such as for an insurance claim or a dispute, submit the original. This is general information, not legal advice.

When the photo is clean but small

If your diagnosis landed on low resolution — the photo is sharp, just too small for the print or screen you have in mind — that's the job Upscale image is built for. The other faults already have their own row in the table above.