Why your enhanced photo looks worse
Almost every disappointing result comes from one of six causes: the wrong tool for the damage, an over-processed face, a bad capture feeding the model, running passes in the wrong order, a source too degraded to rebuild, or a result that is technically sharper but has lost the character that made the photo worth keeping.
You used the general enhance on an old print
This is the most common one by a distance. The general pass sharpens. An old photograph is usually faded, physically damaged and low resolution as well as soft, and sharpening addresses one of those four.
Worse, sharpening a faded print emphasises the flatness. You end up with crisp edges inside a grey, low-contrast image, which reads as harsher rather than better. Run Repair instead, which handles contrast, damage and resolution together.
The face went waxy
Skin rendered as a smooth gradient with no pores, matched glassy eyes, hair that looks painted. This happens when a general model is asked to rebuild a face, because it has no idea which smudges are eyelashes and which are noise.
Use the face tool rather than the general enhance on portraits. If the face tool itself produced the waxy result, the source was probably too degraded, and the honest answer is that a less impressive but more truthful version is the better one to keep.
The capture was the problem, not the app
If you photographed a print at an angle, under a ceiling light, or through glass, the model is working on a copy that already contains glare, shadow and distortion. It treats all of that as part of the picture and enhances it faithfully.
Recapture before you conclude the tool failed. Flat surface, soft daylight, no flash, phone parallel to the paper. This one change fixes more disappointing results than any setting.
You ran the passes in the wrong order
Sequence matters because each pass works on whatever the previous one produced. Repair first, so damage and fading are gone. Face restoration second, once the model is not trying to rebuild an eye through a crack. Colorize last.
Colorizing before repairing is the classic mistake: colour applied over a crack makes the crack more obvious, because the damage now has a colour of its own that does not match anything around it.
There was not enough left to work with
A photograph of a photograph, a heavily forwarded chat image, a face forty years faded past recognition. In these cases the model is generating rather than restoring, and generated detail is exactly what people mean when they say a result looks fake.
The test is whether a human can identify the subject in the original. If you cannot, no tool can, and anything convincing that comes back has been invented.
It is technically better and emotionally worse
This one is real and rarely discussed. A restoration can be sharper, cleaner and higher resolution while feeling wrong, because the grain, the warm cast and the shallow depth of field were part of why the photograph looked like 1968.
Strip those out and you get a picture of people in period clothes rather than a photograph from that period. If a result bothers you and you cannot say why, compare it against the original and look for what was removed rather than what was added.
What to do when nothing helps
Keep the untouched capture and stop. A scan that shows what the photograph actually looks like today is more valuable than a confident reconstruction that satisfies nobody who remembers the subject.
For the one or two photographs that genuinely matter most, a human retoucher working from other pictures of the same person will beat any automated tool, and knows to ask you what your grandfather looked like.
Questions people also ask
Why does AI enhancement make skin look plastic?
A general model smooths what it cannot interpret, and skin texture is mostly fine detail it reads as noise. Use a portrait-specific tool, and prefer the result that keeps some texture even if it looks less dramatic.
Can I undo an enhancement?
You do not need to. The enhanced image saves as a new file and your original stays in your library untouched, so you can simply delete the version you dislike.
Related guides
How AI photo restoration works
How AI photo restoration rebuilds faded, blurry and damaged pictures, why the results are plausible rather than recovered, and when to trust them.
Read the guide →
How to photograph old photos with your phone
Capture old prints properly with a phone camera. Light, angle, glass, glare and focus, and why the capture puts a ceiling on every restoration that follows.
Read the guide →
How to fix old photos
Fix old photos by working out what is actually wrong with them. A five-point check for fading, blur, damage, resolution and colour, and the fix for each.
Read the guide →