Skip to content
GPU-assistedUpload required

Repair Marks on a Damaged Print

The model looking at your upload does not know it is looking at a print. It sees one flat picture in which some lines are wires and some lines are creases, some patches are shade and some patches are water stains, and nothing marks which is which. Everything this page does well and everything it gets wrong follows from that single fact, so it is worth reading before you decide what to send it.

Use this without the search next time. Prathom Workbench puts Prathom's tools in your toolbar.

Add to Chrome — free

Drop an image here, or click to browse

Up to 25 MB. Uploaded temporarily, deleted after processing.

GPU-assisted — results vary slightly between runs.

What it does

  • Creases, cracks, stains, and abrasion targeted
  • Surface rebuilt from the picture around it
  • Composition and identity held in place
  • One megapixel working size

How to use Restore Damaged Photo

  1. 1

    Scan the print flat if you can

    A crease under raking light casts a shadow as well as showing a crack. Under glass on a flatbed only the crack remains, which is much less to repair.

  2. 2

    Run the repair

    The picture is repainted with the instruction to eliminate surface marks and rebuild what was underneath them, keeping people and objects unchanged.

  3. 3

    Look at the busy areas first

    Patterned fabric, brickwork, railings, and foliage are where an unwanted repair shows. Straight lines in the picture are the usual casualty.

How it works

Your image is scaled to about one megapixel and repainted by an image editing model at a moderate denoise with a fresh seed each run. The instruction names the damage types directly — creases, cracks in the emulsion, water stains, foxing, abrasion — and asks for the picture underneath them to be rebuilt, while the negative prompt argues against leaving crack lines behind and against smearing the repair into a blur.

The denoise sits higher than the faded-print page uses, because a mark has to be eliminated rather than merely brightened, and lower than the torn-print page, because the material is still there.

The model cannot see the print

This is the mechanism behind every good and bad result the page produces.

You are holding an object: paper, an emulsion on top of it, and damage to both. The model receives a flat array of pixels in which the damage is not marked as damage. A crease appears as a light line with a dark line beside it. A crack appears as a fine dark thread. A water stain appears as a soft brown region with a harder edge.

The trouble is that ordinary photographs are full of light lines beside dark lines, fine dark threads, and soft brown regions with hard edges. Kite string, door frames, a scar, a shadow under a roof, a bald patch, a stripe on a shirt. The model's only handle on which is which is plausibility, and plausibility is a statistical judgment about pictures in general, not a reading of your particular one.

So the practical rule falls out on its own: the more your photograph looks like damage, the more of it you should expect to lose.

What repairs well

Damage over a plain or smoothly varying region is close to a solved problem. A crease across a wall, a sky, a face, a floor, or a shirt front is rebuilt convincingly, because the surrounding pixels genuinely determine what belongs under the mark. Foxing spots are the easiest of all: small, isolated, and surrounded on every side by intact picture.

Damage over structure is where it struggles. Repeated patterns are the worst case, because a pattern has a period and a phase, and a stain wide enough to hide several repeats leaves nothing to lock onto. Brick courses shift, stripes change width, railings gain or lose a bar. It looks right at a glance and wrong when you count.

The scan matters more than the settings

Two prints with identical creases can produce very different results depending on how they were captured.

Photograph a creased print at an angle to a lamp and the fold casts a shadow along its length, so the file records both the crack in the emulsion and a broad tonal band that has nothing to do with the picture. That band is a second, larger, softer piece of damage and it is harder to remove than the crack.

Put the same print on a flatbed under the lid and the paper is pressed flat and lit from directly along the surface. The shadow largely disappears and only the physical crack remains. That is a smaller job with a better outcome, and it costs nothing but the choice of device.

Publication gate

This page ships once the workflow has been run against a creased print, a water stained print with a patterned background, a foxed print, and a photograph containing wires and railings with no damage at all, checking in that last case that nothing was repaired away.

Examples

Print creased down the middle

wedding.jpg - 3000x2100, a fold line through the group
prathom-restore-damaged-photo.png - 1195x836

The intended case. A crease across faces and plain background rebuilds well because the surrounding content tells the model what belongs.

Photograph with a water stain over a striped wall

hall.jpg - 2400x1800, brown tidemark across a striped wall
hall-repaired.png - 1155x866

Mixed result. The stain goes, but the stripes are frequently rebuilt at the wrong spacing, because their period cannot be read through the stain.

Frequently asked questions

Why did it remove something that was part of the picture?

Because to the model a crease and a genuine dark line are the same object. A telephone wire, a window frame edge, a scratch on a wall, a parting in someone's hair — all of these look like the marks it was asked to eliminate. It has no access to the print itself, only to a picture of one, so it cannot use surface as a clue.

Will it work on a photo stuck to glass?

Only for what survived. When a print is peeled off glass the emulsion often stays behind in patches, so the image is physically gone in those areas and this fills them with invention rather than recovery. Do not peel it — scan it through the glass first, and treat any output as a reconstruction.

What are those small brown spots?

Foxing, usually. It is age spotting in the paper base, driven by humidity and often by impurities or mould, and it grows over time in storage. This clears it well because the spots are small and the surrounding picture is intact, which is the easiest case the workflow gets.

Should I repair the scan or the print?

Repair the scan and never the print. Physical retouching, tape, glue, and trimming are irreversible and destroy value in an original that may matter to other people later. Scan it first, keep the scan, and do all the work on a copy of that file.

Is the AI Restore Damaged Photo tool free, and do I need an account?

Yes. The AI restore damaged photo is free, and there is no account to make first. Nothing is billed, no email is asked for, and no watermark is added to what you download. Uploads are capped at 25 MB, the work is queued on a shared GPU host rather than run in your browser, and finished jobs are held for thirty minutes before they are cleared.