You run your grandparents' wedding photo through a restoration app and the result is amazing: the crease is gone, the face is sharp, the faded color has turned to color. The first reaction is always the same. The second, when someone in the family looks at it, is usually: "his nose wasn't like that before".
The issue here is the gap between what is impressive and what is faithful, and it stems from how the technology works.
Recovering is not the same as rebuilding.
There are two very different operations happening under the same button.
Recover It's about working with information that's already there. Adjusting the contrast of a washed-out photo, correcting the yellowish tint of aged paper, removing the dust that was scanned along with it. The data exists in the file—the program just organizes it better. Here, the result is faithful.
Rebuild That's something else entirely. When a photo is blurry or has a piece missing, the information no longer exists anywhere. What the model does is look at millions of photos it has already seen and fill in the gaps with what it needs. usually Being there. It's not a memory of your grandfather. It's statistics about faces in general.
Hence the phrase that sums it all up: The result is plausible, not true.. And the worse the original, the more the program improvises — because there's less left to salvage.
Where does this appear most often?
Some details frequently give away the reconstruction:
- Eyes and teeth. These are the areas that the model "corrects" the most, and those that most change a person's identity.
- Skin texture. It usually comes out too smooth, looking like a modern portrait in a photo from sixty years ago.
- Written in the background. Sign, banner, lettering — the model fills in shapes that look like letters but don't form any words.
- Hands and fingers. They remain the classic weak point.
- Color. The colorization is always an educated guess. The dress could have been a different color, and there's no way the program could know.
If the photo has documentary value—to prove what someone looked like, to identify a person, to serve as a record—these points matter a great deal. If it's just to hang on the wall and evoke emotion in the family, they matter less.
The decisive step: digitization.
This is where almost everyone loses quality before they even begin, and no app can recover from that afterwards.
Taking a photo with a cell phone works, as long as you avoid the classic mistakes:
- Diffused light, never flash. The flash creates a glare in the center that completely obscures information. Near a window, on a clear day without direct sunlight, is the best setting.
- Cell phone parallel to the photo. Photographing from the side distorts the geometry, and the program will "correct" the face based on the distortion you created.
- Without its shadow overhead. It seems obvious, and it's the most common mistake.
- Flat photo. If it's curled up, leave it under weight for a few days before
If the photo is behind glass or plastic, remove it. The reflection enters the image and becomes part of what the program tries to interpret.
Those who have access to a standard scanner have a real advantage: the light is uniform and the geometry is perfect. It's worthwhile for the photos that truly matter.
The correct order to stir
Trying to do everything at once usually yields worse results than doing it in stages.
- Cut and straighten First of all, let the program work only on what matters.
- Correct the light — contrast and color dominance. Many "damaged" photos are simply yellowed.
- Only then use the automatic restorer. With the original already adjusted, he needs to reinvent the wheel less.
- Colorize last., if applicable, and always keeping the black and white version.
And the rule that matters most of all: Never overwrite the original file.. Keep the raw scan in a separate folder. It's the negative of the story—you can redo the processing in five years with better tools, but you can't recover what was erased.
Before uploading your family photo
Most of these services process the image on their server, not on your device. In other words, the photo is taken from your phone.
It's worth opening the privacy policy and looking for two things: how long the file is stored and whether it can be used to train the company's models. The answers vary considerably, and free services tend to be more permissive than paid services—which makes sense, because something has to pay the bill.
For document photos, photos of children, or any sensitive situation, a tool that processes photos within the device itself is the most reliable choice, even if it delivers slightly inferior results.
When retouching turns a photo into a completely different one.
There's a point where restoration stops being restoration. If the result changes the shape of the face, the expression, or the eye color, you no longer have a photo of your grandparents—you have an illustration inspired by it.
There's nothing wrong with wanting this, as long as it's done consciously. The important thing is to be careful when sending it to family: it's worth mentioning that it underwent automated processing. This prevents someone twenty years from now from mistaking the reconstructed face for a record of what the person looked like.
Summary
These apps are very good at fixing what time has done to paper—stains, creases, fading—and only convincingly at recreating what has already been lost. Scan carefully, work in stages, keep the original untouched, and treat the reconstructed face for what it is: a good approximation.