How to Unblur and Restore Old Passport and ID Photos

When my great aunt passed away, we found her mother's old passport from 1921 among her things. The photo was about one inch square. It showed a woman in her twenties with dark hair and a serious expression. It was also blurry, faded, and had a crease running diagonally across her face from being folded inside the passport for a century. It was the only known photo of my great grandmother.

Passport and ID photos are often the only surviving images of ancestors from the early twentieth century. They are always small, usually low quality, and frequently damaged. But because they are formal portraits shot in controlled lighting, they contain good underlying facial data. Here is how to pull it out.

Start with the best possible scan

The photo is tiny. You want to capture every bit of detail that still exists. Scan at 1200 DPI if possible. At that resolution, a one inch photo gives you 1200x1200 pixels. At 300 DPI, the same photo gives you only 300x300 pixels. That four times difference in resolution is the difference between an AI that can reconstruct a face and an AI that is guessing from almost nothing.

If you do not have a scanner, photograph the passport photo with your phone. Use natural daylight near a window. Hold the phone parallel to the photo. Fill the frame with the photo as much as possible. Tap to focus on the eyes. Take multiple shots and pick the sharpest one.

Isolate the face from the passport background

Old passports have stamps, watermarks, serial numbers, and decorative borders that overlap with the photo. These confuse the AI. Before restoration, crop the photo tightly to just the face area. If the passport text or stamp overlaps the face, crop as close as you can without cutting into facial features.

For my great grandmother's passport, the photo was surrounded by an embossed government seal that bled into the edges of the portrait. Cropping removed most of it. What remained was faint enough that the AI ignored it.

AI restoration of tiny faces

The face in a passport photo is usually between 20x20 and 60x60 pixels at typical scan resolutions. At that size, you can tell it is a face but you cannot see details. AI face restoration models like GFPGAN are trained specifically for this. They analyze the pixel pattern, identify facial features, and reconstruct them at higher resolution.

For my great grandmother's photo, the original face was roughly 40x50 pixels after cropping. The AI face restoration pass recovered her eye shape, the bridge of her nose, and her mouth. The crease across her face was filled in. The result was a 400x500 pixel portrait that looked recognizably like a person rather than a blurry stamp-sized photo.

I then ran a 4x upscale to bring the restored portrait to 1600x2000 pixels. That is large enough for a 5x7 print at 300 DPI. From a one inch passport photo to a framed portrait. Cost on ClarifyPix Face Restoration was 4 credits for face restoration and 4 credits for the 4x upscale. Eight credits and about fifteen seconds of processing for a portrait that did not exist before.

Managing expectations

The restored portrait will look like the person, but it will not look like a modern digital photograph. The AI is reconstructing facial features from limited data. It gets the broad structure right. Eye shape, nose position, mouth width. It does not capture every fine detail that a real photograph would have. The result is a recognizable likeness, not a perfect reproduction.

My mom looked at the restored portrait of her grandmother and said it looked like her. She had no other photo to compare it to, but the face matched her memory of the woman who raised her. That is about as good a validation as you can get for AI restoration of a hundred year old passport photo.