Repair damaged old photos — creases, tears and scratches
The AI identifies what is missing and fills it in, without distorting faces
Tap to pick from your library — results in under a minute
A photo that spent forty years at the bottom of a box usually carries three kinds of damage: the white line left by a fold, a worn-away corner, and mould spots from damp. What they have in common is that the original information is genuinely gone — repair is not erasing the mark but working out from the surroundings what should have been there: how the broken collar continues, how the two halves of a face split by a crease line up, whether the missing corner was wall or sky.
Different damage, different handling
"Damage" is a loose word. In practice the type matters, because each leaves a very different amount to work from — and that sets how far the repair can go.
| Type of damage | Usual cause | How far it usually gets |
|---|---|---|
| Fine web-like cracks | Ageing of the emulsion layer | Essentially invisible — thin cracks, intact on both sides, plenty to infer from |
| Creases and fold lines | Stored folded for years | The line itself goes; one running across a face deserves a careful check |
| Tears, split in two | Physically torn | Align the pieces, shoot them as one, and the seam closes convincingly |
| Missing corners, abrasion | Rubbed edges, insect damage | Background fills invisibly; a missing person or object can only be plausibly reconstructed |
| Mould spots, water stains | Damp storage | Shape fills well, but discolouration usually comes with it — correct colour too |
| General scratches and specks | Common on scans and re-shoots | Cleans up thoroughly — the most immediately rewarding case |
The rule that matters most: do not turn them into someone else
Filling in is technically "generate the missing content from context", and generation has a pull — it drifts towards the typical. Where a crease crosses the bridge of a nose, given free rein it will return a nose that is more regular, more symmetrical, and less like the person. So one rule is absolute here: intact pixels are never touched, work happens only inside the damaged area, and it stays conservative.
- Intact regions take no part in generation, so a whole face never gets quietly "improved"
- Where damage crosses features, structure wins — better a faint trace than a rebuilt nose or eye
- Filled areas are matched to the surrounding grain, noise and sharpness, so no suspiciously clean patch remains
- You get a pixel-level comparison slider — whether it still looks like them is a judgement only someone who knew them can make
Four steps to repair a damaged photo
- 1
Prepare the photo first
Reassemble torn pieces and align them; flatten a curled print under books for a day or two. This is the best-spent time in the whole process: a misaligned seam is one more line to guess, and flattening removes every later problem caused by shadow and distortion.
- 2
Photograph or scan it
Lay it flat in daylight, square-on and filling the frame, without flash — a hotspot on glossy paper reads as damage. A scanner is better still, at 300 dpi or above. Upload the original file; do not run it through a photo app first.
- 3
Select the damage types
"Scratches and damage" is already selected for you. Tick yellowing or blur as well if they apply and it all runs in one pass. Add anything specific to the notes, such as "leave the mark in the bottom right, it was there originally".
- 4
Compare, re-run, print
Work through the slider, concentrating on where the damage was and on faces. Add a note and run again if needed (there is a daily free allowance, shown on the page). Then download the full-resolution file or order a 6-inch print to frame.
Why it is worth printing
A repaired photo that only ever lives on a phone can still be lost — a new handset, a cleared camera roll, a forgotten password. A print depends on no device, and genuine inks on professional photo paper hold up in normal storage. That is the point of all this for us: restoration is the middle step, and turning it back into a photograph you can hold is the end of it.
Common questions
My photo is torn in two — can it still be fixed?
Yes. Lay both halves together, align them carefully and photograph them as a single image. Get the edges to meet as closely as you can — an obvious gap is one more thing the AI has to guess. Where too much is missing it infers from the surroundings rather than inventing a face.
Do mould spots and water stains count as damage?
Yes, same pass. But mould usually comes with discolouration — it consumes the dye in the paper, so the affected area often goes pale and yellow at once. Tick "fix yellowing/fading" as well; repairing shape without correcting colour leaves a patch that does not match.
Will repeated repair make the person look less like themselves?
Intact areas are not touched at all — only damaged spots are filled, and that is the first rule here. Check with the comparison slider afterwards, especially where a crease runs across a face. If it is not right, add "scratches only, leave the face alone" to the notes and run again.
Can a large missing area — a whole corner, half a body — be filled in?
It depends what is missing. Background — wall, sky, ground, the continuation of clothing — usually fills invisibly. A face, a hand, or anything unique in the frame leaves nothing to work from, so what comes back is a plausible version rather than the original. In those cases we prefer to stay conservative: better a faint trace than a person conjured from nothing.
The photo is covered in fine cracks like a spider web — what about those?
That is the emulsion layer ageing, it is very common, and it is the most rewarding case: the cracks are thin, randomly oriented and intact on both sides, so there is plenty to infer from and the result is usually very clean.
Should I touch it up in a phone app before uploading?
Please do not. Any pre-processing loses information: sharpening hardens crack edges and makes them harder to fill, compression adds fresh noise, and auto-enhance often flattens faint detail away entirely. Upload the largest, most original file you have.
Can I print it afterwards, and at what size?
Yes, straight from the result page. 6 inch is the safe choice for a repaired photo; 8 inch is fine when the original is sharp and the damage was limited. Printing uses the unlocked full-resolution file, not the preview you see on screen.
How many times can one photo be repaired?
You can re-run it; each run differs slightly. There is a daily free allowance, shown on the page. If one area did not fill convincingly, adding a line to the notes and re-running usually sorts it.