Task · Restoration
Copy-ready old photo restoration prompts for repairing faded, scratched, or blurry photos — written to restore what is there, not invent what was lost. AI restoration is a copy, not an archival recovery: keep your original scan.
№ A1 · Task · restoration
A good old photo restoration prompt names the damage before it names the fix — scratches and fading first, the family face always held. This small set of all-round prompts covers the most common prints, faded, torn, or freshly scanned, and each one restores damage while holding the original grain.
№ A2 · Task · workflow
To restore old photos with ChatGPT, scan flat, describe the damage in plain words, restore in one conservative pass, then correct anything the model guessed in a short follow-up.
№ Identity · Keep the family face · two safeguards
A restored photo only matters if the person in it is still recognizable. The image model rebuilds the whole picture, so faces can drift toward a generic likeness. These prompts ask the model to hold identity — they reduce drift, they do not guarantee a perfect match, so always compare against the original.
Name identity as a hard constraint
Before describing the repair, tell the model what must stay: “Preserve the exact face — bone structure, eye shape, and expression. Repair only damage; do not re-draw features.”
Repair around the face, not through it
Heal scratches and fading in the background, clothing, and edges first. Touch a face only where damage clearly crosses it, and keep that repair minimal so the likeness your family remembers stays intact.
№ Honesty · Restore, do not reconstruct
AI restoration is not archival restoration. To fill missing or damaged areas the model guesses from what it has learned, which means it can add a different earring, an imagined sleeve, or a plausible-but-wrong feature that was never in the original. Keep the untouched scan and review every added detail.
Restore only what is visible
Constrain the edit explicitly: “Restore only what is visible in the photo. Do not add jewelry, clothing, background, or facial features that are not already there.”
Treat a missing area as missing
When a corner is torn away or a face is half-gone, a reconstruction is a guess, not a recovery. Decide whether to leave the gap honest or to accept an interpretation — and label it as one when you share it.
№ FAQ · Old photo restoration · frequently asked
It can. Even a careful photo restoration prompt can’t change that — AI restoration is not archival restoration. To fill damaged or missing areas, the model guesses from what it learned — which means it can add a different earring, an imagined sleeve, or plausible-but- wrong facial features that were never in the original photo.
The avoid-invented-details section above has copy-ready wording that constrains the model to restore only what is visible.
You can ask it to — name identity as a hard constraint (bone structure, eye shape, expression) and limit the edit to repairing damage. The preserve faces section above explains how.
This reduces drift; it does not guarantee a perfect likeness. Because the model rebuilds the whole image, always compare the restored face against the original scan before sharing it with family.
Only if you accept that the colour is an interpretation, not a record. Unless the true colours are documented, the model is guessing — so ask for restrained, plausible colour rather than confident, saturated guesses.
A valid choice is to repair the damage and keep the photo black-and-white, leaving the era’s own look intact.
No. Treat the restored image as a shareable copy, never a replacement. AI restoration regenerates the picture rather than recovering the original emulsion, so it is not an archival-grade process.
Always keep the untouched scan or print. If the photo has real historical or family value, a professional conservator is the right path for the original.
