Old Photo Restoration

Prints wear in predictable ways, and most of it can be repaired digitally. Restoration works on the scan, so your original print is not touched.

Restoration uses a generative image editing model. Instead of only filtering pixels, it understands the scene and fills damaged areas with content that matches what is around them. This is what lets it remove a crease that runs across a face or a sky.

The same ability is a risk: a generative model can also redraw things it should have kept. We prompt the model to keep every face, expression, pose and item of clothing as it is, and not to add or remove people or objects. In our tests on a scratched historical portrait, this approach cleaned the damage and stayed closer to the original than face-restoration tools, which left scratches in place and smoothed skin.

It is still worth checking. Always compare the restored face with your original using the slider. If an expression looks different, run it again, or use a lower-damage version of the scan.

If you want to keep the historical look, choose restore only. The photo stays in black and white or sepia and simply looks like it did when it was new. If you want the picture to look alive to the family, choose restore and colorize. Doing both in one pass is better than colorizing first, because color applied to a crease or stain makes the damage more visible and harder to remove afterward.

Most disappointing results start with a poor scan. Remove the print from its frame or album if you safely can, wipe dust with a soft dry brush, and scan flat at 600 DPI in color mode, even for black and white prints, because the extra channels help the tool see stains. If you must photograph the print, do it outdoors in shade or by a bright window, holding the camera directly above it to avoid perspective and glare.

Keep a copy of the raw scan. It is the master file, and any restoration, ours or anyone else's, is a version of it.

An automatic tool is fast and inexpensive, and it handles most family prints well. Where large parts of a face are missing, where a print is torn into pieces, or where a photograph has legal, museum or commercial importance, a human retoucher working from the original is the better choice. AI reconstructs missing areas by inference, and inference is not recovery.