How to Restore a Torn Photo: What AI Can Fix and What It Can’t

A torn photograph is two problems wearing one coat: a damaged physical object, and a damaged image. Most guides skip straight to software and quietly lose the first half. The order matters, because a tear you handle wrong before scanning becomes permanent in every copy you ever make. Here is the full sequence — physical first, digital second — and an honest map of where AI inpainting genuinely ends.

Can a torn photo actually be restored?

Yes, in the great majority of cases — provided the tear is scanned correctly and the missing area is small. A clean tear with the pieces still present is close to the best case digital repair has: the image information exists on both sides of the crack, so software only needs to close a thin line. A tear that lost material (a missing corner, a chewed edge, a hole) is a different job: everything inside the gap must be reconstructed from surrounding context, and past a certain size that reconstruction stops being recovery and becomes invention.

The practical threshold we use, and the one most conservators describe in similar terms: when a missing area covers roughly 30 percent or more of a face, no software can genuinely recover what was there — it can only draw a plausible substitute. Below that, on eyes, lips, clothing, backgrounds, modern inpainting is remarkably good. Above it, the honest options are a human retoucher working from other photographs of the same person, or leaving the loss visible.

Should I tape the pieces together before scanning?

Only with the right materials, and only if the print is stable enough to handle. Ordinary household tape is one of the most common ways families destroy photographs: its adhesive yellows, migrates into the emulsion, and becomes impossible to remove without solvents that damage the image. The archival alternative is acid-free, lignin-free document repair tape or Japanese tissue with wheat starch paste — the standard of paper conservators (the tissue-and-paste repair is reversible, which is the whole point).

Before you touch anything, three rules:

If the print is valuable, fragile, or the tear crosses a face, the correct answer is: do not repair the object yourself. Photograph the fragments for reference, store them flat in an acid-free folder, and let a photographic conservator handle the physical piece (the American Institute for Conservation maintains a directory). You can still restore the image digitally from your photograph of the fragments.

How do I scan a torn photo so the repair works?

The scan is where most tear restorations are won or lost, and the winning move is counterintuitive: scan the pieces separately, face up, on a neutral background, rather than taping them into a jigsaw first. Lay each fragment on a matte mid-gray or black surface with a small gap between pieces, scan at 600 dpi (higher, 1200 dpi, only for tiny originals or severe damage), save a TIFF master, and turn every automatic scanner enhancement off — auto color, dust removal, unsharp mask. Those corrections are destructive and they guess wrong on aged prints.

Once you have the fragments as one flat scan, aligning them is a five-minute job in any editor with layers — or you can skip alignment entirely and let the restoration model close the seams, which works well when the tear lines are thin and the fragments are already roughly in place. Our digitizing guide covers the scanner settings in full.

What does AI inpainting actually do to a tear?

Inpainting fills a damaged region by predicting what belongs there from the pixels around it. On a thin tear line — a crack a few pixels wide crossing a coat, a sky, a wall — the prediction is essentially perfect, because both sides of the line carry the true image and the gap is narrower than the texture it crosses. On creases and fold lines it is nearly as good. This is why tear repair is one of the tasks where modern tools look like magic.

The physics changes as the gap widens. Inpainting has no memory of your grandmother; it has statistics about what photographs usually contain. Across a missing wedge it will produce something coherent — a plausible continuation of a dress pattern, a believable stretch of lawn — and coherent is not the same as true. Three honest boundaries follow from that:

  1. Repeating texture survives; unique detail does not. Fabric, foliage, sky, brickwork reconstruct convincingly. A specific brooch, a regimental badge, the exact pattern of a wallpaper border do not — what returns is an average of similar things, not your object.
  2. Faces are the hard case. A tear line across a cheek closes cleanly. A missing section of a face does not "close"; it gets replaced by a statistically typical face region. That is why the 30-percent-of-a-face threshold above is the line between restoration and portraiture-by-committee.
  3. Edges of the loss matter more than its size. A hole in the middle of a uniform jacket with intact collar and buttons on either side reconstructs far better than the same hole at the boundary between face and background, where the model must decide which side it is continuing.

None of this is an argument against using AI on tears — for the typical family print it is exactly the right tool, at cents per photo and in seconds. It is an argument for knowing which result you are looking at. A faithful repair removes the damage and leaves the evidence of the photograph alone; an over-reaching one hands you a nicer picture of a person who never stood there. Our pixel-by-pixel face accuracy test measures that difference on real scans, including two failures we found in our own pipeline and then fixed.

When is a tear beyond digital repair?

Four situations, in order of how often we see them:

SituationWhat software can doHonest verdict
Clean tear, all pieces presentCloses the line, recovers full imageRoutine. AI first, judge at 100% zoom
Small missing chip (corner, edge, < a thumbnail)Reconstructs background or clothing convincinglyGood result; treat reconstructed area as interpretation
Missing area touching or covering part of a faceDraws a plausible substitute, not the original personUsable only if you accept invention; else human retoucher with other photos of the person
Large loss (> ~30% of a face) or shredded fragmentsProduces a coherent strangerNot restoration. Conservator for the object; consider leaving loss visible in the image

The last row deserves a sentence vendors avoid: sometimes the respectful choice is to keep the wound. A tear across a wartime portrait is part of that object's history, and a restoration that erases it erases evidence. Many families print both versions — the repaired image for the wall, an unretouched scan for the archive. Neither choice is wrong; choosing blindly is.

How do I keep the repaired file from becoming the new original?

Keep three things separate and you can never lose the evidence: the raw scan (TIFF, untouched, tears and all), the restored file (JPEG for printing and sharing), and a side-by-side for the family album so nobody later mistakes the repair for the record. Store the masters in two places on two media with one offsite — the 3-2-1 rule — because a restored file you cannot re-derive is just another single point of failure. Our preservation guide covers materials and environment for the physical print, which keeps deteriorating whether or not you scanned it.

And start with the hardest photo in the box, not the easiest, when you test a tool: any software flatters a lightly damaged snapshot, and a tear across a face is the test that tells you what you actually bought. PhotoUnfade gives you three free restorations — no account, no credit card, full resolution, with the original always beside the result so you can drag the slider and see exactly what changed and what did not. If your tear falls in the last row of the table above, our AI versus professional comparison explains what to hand a retoucher, and our water damage guide covers the other urgent case families find in the same shoebox. Most torn prints carry a second injury around the tear — a yellow cast or faded contrast from decades in a drawer — and that half of the job is color, not geometry: our yellowing and fading guide covers it, while the step-by-step restoration guide puts the whole sequence, tear included, in one place.

Ready to restore your photos?

3 free restorations, full resolution, no watermark, no credit card.

Start free →

Keep reading

How to Restore Old Family Photos with AI (Step-by-Step Guide)

A complete guide to restoring faded, scratched and damaged family photos using AI — what it genuinely fixes, the over-restoration mistake almost every tool makes, how to choose between subtle and aggressive modes, and how to keep the result looking like your relatives instead of strangers.

How to Fix Water-Damaged Photos (Before They’re Lost Forever)

Water damage is time-sensitive: wet photos can fuse permanently within days and mold starts in 48-72 hours. What to do in the first 48 hours, what never to do, how to save stuck-together prints, and how AI repair handles tide lines, bleeding and mold.

AI Photo Restoration vs. a Professional Retoucher: Honest Comparison

When is AI restoration enough, and when should you pay a human? Cost, quality and turnaround compared honestly — including the cases where AI invents details that were never there, and how to tell a faithful repair from a repaint.

How to Digitize Old Photos: Scanning Settings That Actually Matter

The scanning settings that determine whether your digital copies are archival quality or throwaway — DPI, file format, color depth, and the three mistakes that ruin a family photo archive before it starts.

How to Fix Yellowed and Faded Photos (And Why They Changed Color)

Yellowing and fading have different causes and need different fixes. What is chemically happening to your old prints, why you cannot safely bleach them, and how to restore the color digitally without making people look gray.

How to Preserve Old Family Photos So They Stop Fading

The four things destroying your old photographs, the storage materials that are actually archival (and the ones marketed as safe that are not), and how to prioritize a shoebox you cannot fix all at once.

Do AI Photo Restorers Change Faces? A Pixel-by-Pixel Test

We measured whether AI photo restoration changes skin tone, invents color in black-and-white photos, or repaints faces — including two failures we found in our own pipeline. Real numbers, real methodology, and how to run the same checks yourself.

Best AI Photo Restoration Tools in 2026: An Honest Buyer’s Guide

There is no single best AI photo restoration tool — there is the right tool for your photos. A plain-language guide to the four categories, what each is good at, the subscription traps to avoid, and how to test any tool in five minutes.

How Much Does Photo Restoration Cost? A 2026 Price Guide

What one photo really costs to restore in 2026 — AI services, freelance retouchers, studios and conservators compared, the arithmetic for a 200-photo shoebox, the hidden fees to watch for, and our own measured cost per restoration including the plan on which we lose money.