AI Photo Restoration

Restoration asks for three jobs at once, and the risk is that all three get done enthusiastically. The instruction above eliminates scratches, dust, and tears; sharpens detail to an exceptional degree; and corrects faded or discoloured colour toward something vibrant and natural. It works inside the original frame with the subject kept in view, so the fair way to judge a result is against the scan you started with, not against your memory of the person standing in it.

Decide which job your photograph needs

A print with one long tear across a corner does not need the whole frame sharpened, and a colour print that has simply shifted magenta does not need its contrast rebuilt. The parts cannot be switched off individually here, but they can be biased, because the prompt box takes description of the subject, style, composition, and important details. Asking for the grain of the paper print to remain, or for the original muted palette to be kept, pushes back against the defaults without fighting them.

Reconstructed is not recovered

Where a tear has taken a piece of a face, the pass invents a face to fill the gap, working from what surrounds the loss and from what a person normally looks like. That is genuinely useful for a shoulder, a lapel, or a hedge behind a garden party, and it is precisely why eyes, teeth, fingers, and handwriting deserve a close read. Detail that did not survive the damage is being argued for rather than retrieved, so keep the original scan alongside the result, and keep the print itself if the photograph matters to anyone.

Faces, textures, and small text at the size you mean to use

The checklist this page gives you is edges, faces, textures, and small text, checked at the intended output size. Faces go waxy when clarity is pushed too far, since skin loses its pores and the eyes flatten. Woven cloth, tweed, brick, and grass either gain plausible detail or turn into smear. Small lettering is the harshest test: an address on a shopfront, a date, a number on a jersey often comes back changed rather than corrected. Look once at full zoom, then at the size it will actually be printed or posted, because over-sharpening hides at thumbnail scale.

Frequently asked questions

about restoring old photographs

Will this bring back detail the damage destroyed?

Not reliably. Missing areas are reconstructed from their surroundings, so a scratch across an eye is filled with a plausible eye rather than the photographed one. Compare every damaged feature against the scan before you print a result or hand it to a relative.

Why does the skin in my restored photo look waxy?

That is the sharpening and clarity reading badly on skin, which is the texture least forgiving of reconstruction. Ask in the prompt box for natural skin texture and the grain of the original print to remain, then judge at full zoom.

Do I have to clean the scan first?

No, since dust, scratches, and tears are named in the request. What is worth checking afterwards is the difference between debris on the scanner glass and real damage on the print, because a speck from the glass sits on top of everything, including the subject.

Can I keep the faded look instead of correcting the colour?

Say so, because the default is vibrant, natural, pristine colour, which lifts a faded print considerably. If the muted palette is part of why you like the photograph, describe the tonal look you want preserved.