Restore Old Photos in Seconds

An image model does what a retoucher used to do by hand. Trained on millions of photographs, it knows what a scratch, crease or faded highlight should have looked like, paints the detail back in, sharpens soft faces and — if you ask — colorizes a black-and-white print, all in a few seconds.

You can try it without an account. After that each restore costs 2 credits, and signing in also gives you higher-resolution output and a history of everything you have restored.

Google Nanobanana stays close to the original, which suits lightly faded photos and faces you want to keep recognisable. ChatGPT Image rebuilds more aggressively and copes better with heavy damage, but drifts further from the original. They cost the same, so on an important photo it is worth running both and keeping the better result.

JPG, PNG, WEBP and HEIC, up to 10 MB. A flat, evenly lit scan gives the best result, but a straight-on phone photo of the print works too — just avoid glare and hard shadows across the face.

Small tears, creases and stains are restored reliably. For large missing areas the AI reconstructs plausible detail rather than the true original, so results vary — try ChatGPT Image for heavier damage.

Colorization is an educated guess based on what the model has learned about skin, fabric and scenery. It looks natural, but the exact shade of a dress or a wall cannot be recovered from a black-and-white photo.