How to fix a blurry photo: camera shake or out of focus?

Two kinds of blur: camera shake and missed focus

Before you try to fix a blurry photo, work out which of the two blurs you are looking at. They come from different mistakes and they respond very differently to repair.

Motion blur (camera shake). The camera, or the subject, moved while the shutter was open. Every point in the scene was dragged along the same path, so edges are smeared in one direction and crisp in the other. A shaky handheld photo at night usually shows a short diagonal streak; a child running through the frame shows a horizontal one while the wall behind stays sharp. The information is still there, spread along a line.

Out of focus (defocus). The lens focused on the wrong plane: the wall behind the person, the glass in front of them, or nothing at all because the phone had not finished focusing when you tapped. Every point becomes a small disc, about the same in all directions. Nothing is dragged, everything is softened evenly, and the softness grows with the distance from the plane that was actually in focus.

The quickest way to tell them apart is to find a small, bright point of light and look at it at 100 %, one screen pixel per image pixel, not fitted to the screen. Good candidates: the catchlight in an eye, a street lamp, a candle flame, the standby LED on a television. Then read its shape:

• A line or elongated streak with two ends: the camera moved. The length of the streak is the amount of motion blur in pixels and its angle is the direction the camera went.
• A round or polygonal disc with fairly even brightness and sometimes a brighter rim: the shot is out of focus.
• A double point, two clear copies of the same light a few pixels apart: the camera jumped (a hard tap on the screen, a shutter bounce). This is the hardest case, and we come back to it below.

You will often find both at once, a slightly missed focus plus some shake, and the streak then has fuzzy ends. Treat it as whichever is bigger: a long streak with soft ends is a motion problem; a wide disc with a small smear is a focus problem.

No point of light in the frame? Look at a straight edge, such as a door frame, in two directions: one sharp and one smeared means the camera moved; both equally soft means the focus was off.

Which blurry photos can be sharpened, and which cannot

Motion blur is the friendlier of the two. The light from each point has been spread along a line, and a model trained on exactly that kind of smear can pull it back together, because the line still carries the information about where the edges were. A streak of a few pixels to a few dozen in one clear direction is where you can genuinely recover a shaky photo rather than just make it look crunchier.

In our tests on 19 September 2026, a 1600 × 1200 px photo with 29 pixels of motion blur was sharpened in 7 seconds. The result is not a different photograph: the edges land where they were and letters that were a smear become legible. A streak of that length, under 2 % of the width of the picture, is a useful mental reference for what moderate shake looks like.

Defocus is harsher. Spreading a point into a disc destroys fine detail in every direction at once, and the finer the detail, the more completely it is gone. Mild defocus, where the eyes are a little soft but you can still count the eyelashes, responds well. Strong defocus, where an eye is a uniform grey smudge and you cannot tell the iris from the white, does not: there are no edges left to rebuild, and anything a tool draws there is an invention, not a recovery.

Our test before deciding whether a photo is worth the effort: zoom to 100 % and look at the most important detail, usually a face, sometimes a sign or a number plate.

• Worth it: you can still see where the edges are, even if they are soft. The outline of the eyelid, the line between the lips, the stroke of a letter. The shapes are recognisable, they just lack bite.
• Borderline: you can tell there is an eye, a mouth or a word, but not its internal structure. Expect a cleaner, more pleasant image, not a sharp one, and keep the strength below maximum.
• Not worth it: the detail is a smooth patch with no structure, the streak is longer than the feature it is dragged across (say a 40-pixel smear over a 30-pixel eye), or you see a double image. The tool will tidy the smear but cannot recover what the sensor never separated.

Two cases where the usual advice is wrong. First, double images, two crisp copies of each edge, are not motion blur in the ordinary sense: they are two sharp exposures on top of each other, and a deblurring model tends to sharpen both copies instead of merging them. Second, enlarging does not sharpen. People often try to fix a blurry photo by making it bigger, which produces a bigger blurry photo. If you need both, deblur first and only then upscale the image.

One more caveat: a picture saved from a chat app was compressed and shrunk before it reached you, and its blur is often compression rather than a lens problem. That is a separate case, covered in what can really be recovered from a pixelated WhatsApp photo.

The cost of overdoing it: halos and plastic skin

Any tool that sharpens a blurry photo makes a trade: more contrast at the edges, and reconstructed texture where it believes texture should be. Push either too far and the result looks worse than the blur, even if it looks sharper as a thumbnail.

The two failure modes are easy to recognise once you know them:

• Halos. A bright line along the dark side of an edge and a dark line along the bright side: a branch against the sky gets a white outline, the edge of a dark jacket gets a glowing rim. Classic sharpening filters (unsharp mask, the clarity slider in phone editors) do this by design, which is why they never fix camera shake: they do not move the smeared light back, they only exaggerate its boundary.
• False texture. Skin develops a fine, even grain that looks like sandpaper; hair becomes a bundle of identical hard lines; a plain wall gets a pattern it never had. This is the model filling the gap between what it can recover and what it was asked to deliver.

The defence is simple and not optional: compare at 100 % before you download. A fit-to-screen view hides both faults because the display is shrinking the image; at one screen pixel per image pixel they are obvious. Look at three places: a high-contrast edge (a face against a bright window), a smooth skin area (a cheek or forehead) and a region that should stay soft anyway (the background). If the background has suddenly become crisp and gritty, you have gone too far, because no lens would render it that way.

If you see halos or grain, do not reach for a denoiser to hide them. Lower the strength of the deblur and run it again. The strength control blends the sharpened result with the original: at 100 % you get the full reconstruction, at 60 % you keep most of the recovered edge definition with far less invented texture. For portraits, 60-80 % usually looks like a photograph rather than a drawing; for signs, documents and buildings you can normally stay at 100 %.

If the photo is both blurry and grainy, typical of a handheld shot at night, treat the noise separately, because a strong deblur sharpens the noise too: either remove the noise from the photo first at a moderate setting and then deblur, or deblur at a lower strength and accept a little grain. The order of operations is explained in how to remove noise from a night photo.

Blurry photo of a face: unblur first, restore faces second

Faces are where the stakes are highest and where the wrong order of steps does the most damage. A deblurring model treats a face as edges and textures, like any other object: it brings the eyelid and lip lines back into place, but it does not know what a human face is and will not make a very soft eye crisp. A face restoration model is the opposite: it knows faces extremely well and can rebuild a sharp eye from a soft one, but given a smeared input it may rebuild the wrong eye, because it is working from what faces usually look like rather than from your pixels.

So the order matters:

1. Deblur first. Run the whole photo through the unblur tool with the correct blur type. This puts the geometry back where it belongs: the position of the eyes, the width of the mouth, the line of the jaw. The face restorer that follows then works from the right proportions, not from a smear.
2. Then restore the faces at medium fidelity. Take the deblurred result into the tool that restores old photos, tick "Enhance faces (AI)" and leave the strength slider near the middle, where it starts. In that tool the slider is the face fidelity: it sets how much of the reconstructed face is blended over the deblurred one. At 100 % the model's face replaces yours and the person starts to look like a very smooth stranger with the right haircut. At 50-60 % you keep the person's own asymmetries, skin and expression and only borrow the sharpness.
3. Judge the identity at 100 %, not the sharpness. Zoom in on the eyes and mouth and ask one question: is this still them? A slightly soft but recognisable face is a better result than a crisp face that belongs to nobody. If the answer is no, lower the fidelity and run it again.

Two warnings. If the faces are small, as in a group photo where each head is under 100 pixels tall, the restorer has little to work from and drifts towards a generic face; the tool tells you when that happens. Add up to three reference photos of the same person taken on other days, which steers the reconstruction towards their real features, and still keep the fidelity moderate. And if the face was strongly out of focus to begin with, remember the criterion from the second section: a restorer can make it sharp, but sharp and accurate are not the same thing. Reference photos and the fidelity slider are covered in how to restore old photos with AI.

A note on chat assistants, since many people now paste a blurry portrait into one and ask for a fix: they regenerate the picture rather than process it, so the eyes come back sharp and the face belongs to someone else. We show the difference in enhancing a photo with an AI chatbot.

How to make a blurry picture clear in Airpic, step by step

Airpic runs in the browser with no account and no subscription. Processing the photo and looking at the result is free; the full-size download follows the rules shown on the tool page. Files are deleted one hour after download and after one day at the latest, they are never used for training, and the GPUs that do the work are in the European Union.

1. Upload the original. Open the tool to unblur a photo and drop in the largest version you have, not a copy that went through a chat app. It accepts JPG, PNG (including transparency and 16-bit), WebP, TIFF, HEIC and AVIF up to 25 megapixels, applies the EXIF orientation and keeps the colour profile.
2. Read the suggestion, then confirm it yourself. As soon as the image loads, Airpic analyses the direction of the blur and shows a hint under the two mode buttons: "Looks like motion blur (smear in one direction): Motion mode" or "Looks out of focus (uniform blur): Defocus mode". It is a good first guess, but a guess: apply the point-of-light test from the first section and pick Motion blur or Out of focus yourself. Each mode runs a model trained on that kind of blur, so the wrong one gives a flat, over-processed look.
3. Set the strength. The slider runs from 25 % to 100 % in steps of 5 and starts at 100 %. Leave it there for a first pass on text, buildings and objects. For portraits, start at 70 %.
4. Process. Click "Process image". With the engine warm the result arrives in seconds (our 7-second measurement above was a 1600 × 1200 px file); from a cold start allow one or two minutes. The bar estimates the time left.
5. Compare at 100 %. The result opens in a before-and-after comparator. Click "View at 100 %" to see real pixels, then drag the divider across a high-contrast edge, a skin area and the background, looking for the halos and false grain from the third section. On a phone, use the "See the before" and "See the after" toggle.
6. Adjust if needed. Halos or sandpaper skin: click "Redo with other settings", drop the strength by 10-20 points and run again. Edges still smeared in one direction: you probably chose Out of focus when it was Motion blur; switch and run again. Everything evenly soft but now harsh: the opposite swap. Each pass starts from the original, not from the previous result.
7. Choose the output format and download. Under the advanced settings you can pick PNG, JPG or WebP; keep PNG if you are about to send the result on to face restoration or upscaling, so no compression is added between steps. Click "Download": the result has the same pixel size as the original.
8. Chain the next tool if you need it. Faces: on to the restorer at medium fidelity, as above. Need it bigger for printing: now upscale the image, because an upscaler fed a sharp photo gives real detail, while one fed a blurry photo gives a bigger blur.

That is the whole process: identify the blur with a point of light, decide whether it is worth it, pick the matching mode, keep the strength honest and judge the result at 100 %. Ready? Sharpen your blurry photo in Airpic.

FAQ

Keep reading