How We Restored a 100-Year-Old Family Photo
It arrived as a scan of a scan: a portrait from 1924, creased down the middle, speckled with silver nitrate stains, and faded to a warm sepia ghost. The woman in it stared past the camera with the quiet confidence of someone who had already lived through more than most of us ever will.
Her great-granddaughter wrote to us with one sentence: "She never got to meet my kids. I'd like them to know her face."
The damage report
- Deep creases running through the face — the photo had been folded for decades.
- Widespread scratches and dust embedded in the emulsion.
- Heavy fading in the shadows, with the highlights blown out.
- Overall softness from a low-resolution scan.
Our restoration pipeline
Restoring a photo like this is never one step — it's a pipeline of specialized models, each fixing one kind of damage:
- De-scratch & clean. The first pass removes scratches, dust, and creases while preserving the original texture.
- Face enhancement. A dedicated model rebuilds facial detail — eyes, hair, and skin texture — using what the rest of the photo tells it.
- Colorization. Since the original was black and white, a color model adds plausible, era-appropriate tones.
- 4K upscale. The final pass quadruples the resolution, so the restored portrait can be printed at any size.
When the family saw the result, nobody said much for a while. Then someone whispered: "That's her. That's really her." That's why we do this.
Tips for scanning old photos
- Scan at 600 DPI or higher — more source detail means a better restoration.
- Scan in color even if the photo is B&W; the scanner still captures tonal detail you'll want.
- Don't retouch by hand first — let the AI see the original damage so it can fill it properly.
- Keep the original safe; a good scan is all the AI needs.
Every photo has a story
We restore hundreds of photos every week, and every single one is someone's whole world. The technology matters, but the moment it produces — a face remembered, a story retold — is the real product.
