← All guides · 2026-10-16 · PhotoUnfade team

Can Blurry Old Photos Be Fixed? What AI Sharpening Really Does

Every shoebox has them: the photo where Grandpa is almost sharp, the wedding shot with a slight double image, the portrait that would be perfect if it were just a little crisper. Blur is the most quietly heartbreaking damage in old photos, because it feels like it should be fixable — the person is right there, just out of reach. The honest answer up front: some blur is fixable, some is only improvable, and some is permanent, and you can tell which is which in about two minutes once you know what to look for. Here is how to diagnose yours, what sharpening actually does and cannot do, where AI crosses the line from recovering detail into inventing it, and when the right choice is to leave the blur alone.

Why are old photos blurry in the first place?

Almost always for one of three reasons — missed focus, camera motion, or aged materials — plus a fourth, modern one: a low-resolution copy. The cause decides the prognosis, so diagnose before you repair. A photo that is blurry because the lens missed is a very different problem from a photo that is blurry because someone photographed it with a phone at arm's length.

Missed focus was the default state of consumer photography for decades. The box cameras and fixed-focus Brownies that recorded most of family life from the 1900s through the 1950s had no focus adjustment at all; anything closer than about six feet was simply soft. Later rangefinders and early autofocus compacts missed regularly in dim light. Focus blur looks even and directionless: nothing in the frame is truly sharp, and edges are rounded rather than streaked.

Motion blur came from slow shutter speeds. Early roll-film cameras exposed at roughly 1/25 to 1/50 of a second — fast enough for a posed subject standing still in daylight, hopeless for a child who turned her head. Motion blur has a direction: highlights stretch into streaks and edges ghost into doubles along the path of movement. It is the hardest blur to repair, because the smearing physically mixes information from neighboring parts of the scene, and un-mixing it is guesswork.

Aged materials cause something that looks like blur but is not. As prints fade and fog — the silver image oxidizing, the paper yellowing, contrast collapsing — edges lose their contrast and the whole photo reads as soft. This is the best possible diagnosis, because the underlying detail is still there, merely muffled. The chemistry of this kind of change is covered in our guide to why old photos turn yellow and fade.

What does sharpening actually do to a photo?

Sharpening does not add detail — it increases contrast along edges that already exist. That single sentence explains both why sharpening works and why it so often goes wrong.

The classic technique is the unsharp mask (USM) — a method older than computers, named for the blurred ("unsharp") copy of the negative that darkroom printers sandwiched with the original to boost edge contrast. The digital version does the same arithmetic: find edges, lighten the light side, darken the dark side. The eye reads that local contrast as sharpness. Nothing new is created; information that is already in the file is simply presented more emphatically. Done well, it puts back exactly the crispness a faded print has lost.

Done badly, it produces the two signatures of over-sharpening: halos — bright or dark outlines hugging edges, measurable as overshoot artifacts — and amplified noise, because film grain and dust have edges too, and sharpening cannot tell grain from eyelashes. This is also why repair order matters: sharpening a damaged scan sharpens the damage. Scratches gain bright outlines, dust spots grow rings. Our scratched photo guide lays out the correct sequence — damage first, tone second, sharpening last — and it is the order our own pipeline runs in.

Can AI really unblur a photo?

Partly — and the honest answer requires separating two very different things that are both sold as "AI unblurring." One recovers; the other invents.

The recovering kind is calibrated enhancement: denoising, contrast restoration, and measured sharpening of edges that genuinely exist in the file. The detail was captured by the camera and muffled by age; software turns the volume back up. This is real, and on fade-type blur it works dramatically well.

The inventing kind is generative deblurring: a neural network that has seen millions of sharp faces and, given a blurry one, paints in what a sharp face "should" contain — individual eyelashes, iris fibers, skin pores — none of which exists in your file. The result can look stunning on a phone screen and is, pixel for pixel, partly fiction. We measured the difference on a popular all-in-one photo editor: it pushed measured sharpness up 101% on a test photo and multiplied strong-edge pixels eightfold. No process can extract eight times the edge information from a blurred original — most of that "sharpness" is manufactured texture. Our guide to why enhancement apps make faces look like strangers shows what this does to the people in your photos, and our pixel-by-pixel face accuracy test has the full numbers.

Our own line is deliberately conservative: enhance what exists, invent nothing. PhotoUnfade applies calibrated unsharp masking after repair — never generative face enhancement — and the calibration is published rather than promised: our two modes raise measured sharpness by +29% (subtle) and +77% (full) relative to the original, while halo-type overshoot artifacts actually drop to 8.6% and 7.4% of edge pixels, from 26.1% in the unrestored original. Sharper than the aged photo, cleaner than the aged photo, and still the same photograph.

How much sharpening is the right amount?

Enough to restore the edge contrast the years took away — and not one notch more. The encouraging part is that the right amount is measurable, not a matter of taste. The cautionary part is that measuring it is easy to get wrong. We know, because we got it wrong ourselves.

Our first sharpness measurement came back alarming: an apparent 48% sharpness loss between the original photo and our AI-restored output. Had it been true, it would have meant the restoration was destroying detail. It was not true. We had compared a 531×384 input against a 1024×736 output — different resolutions — and the resizing itself, not the restoration, produced the apparent loss. Re-measured at matched resolution, the real change was -4% in our conservative mode and +10% in full mode: the AI step essentially preserves sharpness. That correction rebuilt our measurement discipline — every figure we publish is taken at matched resolution — and it is a useful warning when reading anyone's sharpness claims, including ours: ask at what resolution the comparison was made.

With honest measurement in place, the sharpening layer itself was tuned on an offline parameter grid — dozens of unsharp-mask settings applied to the same restored image, no AI calls, each scored for edge strength against halo artifacts. The shipped setting sits where measured sharpness is meaningfully up while overshoot artifacts stay below the level already present in the aged original. "As sharp as possible" was never the goal; "as sharp as the truth allows" was.

Can sharpening fix a low-resolution scan?

No. Sharpening rearranges contrast among the pixels you have; it cannot manufacture pixels you never captured. If your only copy of a photo is a 600-pixel-wide image from a 2007 email forward, or a phone snapshot of a print, no sharpening setting on earth will recover Grandpa's eyes. The fix is upstream: find the physical print and scan it properly — 600 dpi, TIFF, flatbed — which typically yields many times the real detail of a phone photo. Our digitizing guide has the exact settings and the reasoning behind each one.

Resolution also sets a physical ceiling on what any restoration can output, ours included. Our pipeline produces up to 2048 pixels on the long edge; across 30 measured production jobs, the average output was 2.42 megapixels. At the 300-ppi standard for photo printing, that is a true 4×6 print, while an 8×10 lands at about 205 ppi — acceptable at arm's length but below the formal standard, a limitation we publish in our cost and resolution guide rather than hide. AI "upscaling" beyond the captured resolution runs into the same question as generative deblurring: the extra pixels have to come from somewhere, and that somewhere is the model's imagination, not your photograph.

When should you leave a blurry photo alone?

When the blur is part of the truth the photo is telling. This is not a consolation prize; for some photographs it is the correct archival decision.

Three cases. First, evidence: if the photo documents something for family history — a house, a uniform, a storefront — a soft original with an honest caption beats a sharpened copy whose details were guessed. Genealogists keep the untouched scan precisely because it is the trustworthy one. Second, identity: when sharpening a face would require inventing features, the sharpened face is no longer your relative. A slightly soft grandmother is still your grandmother; a crisp stranger is not. Third, period truth: a 1943 snapshot carries the tonality and softness of 1943. Pushed to clinical modern sharpness, it stops looking like the day it remembers. Enhancement should serve memory, not overwrite it.

Practically, the safe workflow is always the same: scan at high resolution, keep the untouched scan forever, and restore a copy. That way every degree of sharpening remains a reversible experiment instead of a one-way door.

The bottom line

Diagnose first: focus blur improves, fade-type blur repairs beautifully, motion blur resists, and low resolution cannot be sharpened away — only re-scanned. Then insist on tools that enhance existing detail rather than inventing it, measured claims over adjectives, and your own eyes at 100% zoom as the final judge. The test costs nothing: PhotoUnfade gives you three free restorations, no account, no credit card, with the original always beside the result and a slider between them. Take your blurriest face, zoom in, and check whether the eyes are recovered or replaced. For the full sequence from shoebox to restored archive, start with our step-by-step restoration guide.