Restore old photos: sharp faces and colour with AI

Three steps: Real-ESRGAN cleans grain and scan artefacts and upscales ×2; GFPGAN detects every face and rebuilds eyes, skin and hair; DDColor adds natural colour if you ask for it.

GFPGAN starts from the real face and refines it; on very small or badly damaged faces it may idealise features. Compare with the original in the viewer before downloading.

Yes: tick “Repair scratches and stains” and paint over the scratches, stains or tears; the AI fills them in before cleaning the photo. Large torn or burnt areas are filled with something plausible, not with what was there.

Yes. Leave “colorize” off to keep the original colours; the rest of the restoration applies the same.

No. They are processed and deleted: 1 hour after download, 24 h at most. No copies, no thumbnails.

When the photo is badly degraded, the normal restorer invents a generic face. If you add a sharp photo of the same person (one is enough; up to 3), the face is rebuilt guided by their identity and we show you up to 4 candidates with their measured likeness so you can choose. It takes a few minutes and the references are deleted just like the photo.

Processing and previewing the result is free, with no account. You get 1 free full-size download every month; after that you pay with credits that never expire, no subscription.

A flatbed scanner at 600 dpi with no glare beats photographing the album with your phone. Each face should be at least 100-150 px in the image: below that there are no features left to restore.

Yes: tick “Straighten automatically” and the tilt is corrected by detecting the lines in the photo. If it can't find a clear tilt, it leaves the photo as it is and tells you.