← All guides · 2026-10-12 · Updated 2026-11-09 · PhotoUnfade team

Why Remini Makes Family Photos Look Like Strangers (and What to Use Instead)

You ran a faded family photo through a popular enhancement app, and the person who came back was sharper, smoother — and subtly not your mother. The cheeks are too even, the eyes too symmetrical, the skin texture of a 25-year-old sitting on a 70-year-old face. If this sounds familiar, you are not imagining it. There is a specific technical reason faces come back looking like strangers, and once you understand it you can spot the tools that will do it — before you pay. Here is why it happens, when generative enhancement is still the right choice, and a five-minute test you can run on any tool, including ours.

Why does Remini make faces look different?

Because it does not restore the face — it regenerates one. Apps in this category use generative face enhancement: a neural network trained on millions of portraits that has learned what a sharp, attractive face "should" look like. When your old photo is blurry or damaged, the missing detail cannot be recovered from the file (the information simply is not there), so the model fills the gap with its best statistical guess — and its guesses lean toward the average face it saw most often in training.

That produces a predictable failure pattern. Blur goes away, but what replaces it is partly invented: skin pores smoothed into plastic, irises redrawn with textbook symmetry, a jawline nudged toward the training-set average. Each change is small. Added together, family resemblance — the thing you actually recognize a person by — drifts. This is not a bug in any one app; it is what generative enhancement is. The model is doing exactly what it was optimized to do: produce a plausible, attractive face. "Plausible" and "your mother" are different objectives.

Conservative restoration works from the opposite philosophy: enhance only the detail that already exists in the photograph, remove damage, and never add features the camera did not record. The result is less dramatic on a thumbnail and far more faithful at full size — which is the size that matters when the photo is the only one you have of someone.

What exactly changes when a face gets over-enhanced?

Four things, in the order people notice them:

1. Skin texture disappears. Real faces have pores, fine lines, and uneven tone. Generative models treat texture as noise and replace it with a smooth, airbrushed surface. On a grandparent's portrait this reads instantly as "wrong," because you remember the texture even if you could never describe it.
2. Eyes get redrawn. The iris pattern, catchlight shape, and eyelid fold are unique to a person — and they are the first thing a generative model hallucinates, because eyes carry huge weight in "does this face look good" training objectives. This is why the stranger effect concentrates in the eyes.
3. Geometry drifts. Cheekbones, nose width, lip fullness shift a few percent toward the training average. A 2% change per feature is invisible individually; across a whole face it is the difference between "Dad" and "a man who resembles Dad."
4. Age gets edited out. Models trained predominantly on young, clear faces quietly rejuvenate: wrinkles soften, jowls lift, age spots vanish. A restored 80-year-old comes back looking 55 — flattering, and false.

How big is the problem? Here is what we measured.

We ran the same scans through our own pipeline and a popular all-in-one consumer photo editor, then compared every result against the original pixel by pixel. The full method and numbers are published in our face accuracy test; the findings that matter here:

• Color invention: on a black-and-white portrait that contained 2.5% colored pixels (scanning noise), the consumer editor's output came back 77.5% colored — it had painted an imagined palette over a monochrome photo. Our pipeline returned the photo at 0.0% colored: still black and white, just repaired.
• Sharpening into fiction: the same editor raised measured sharpness by +101% and produced 8 times as many strong-edge pixels as the original — edges the camera never recorded. Our two modes are calibrated to +29% (subtle) and +77% (full), and our overshoot artifact rate actually drops (8.6% and 7.4% of edge pixels, versus 26.1% in the unprocessed original) because we enhance real edges instead of synthesizing new ones. The full calibration — parameters shipped and rejected — is in our sharpening guide.
• Identity preservation: across our test set, skin-tone hue shifted less than 5° — the face stays the same person, in the same light.

One fairness note, because it matters: we have benchmarked only a handful of consumer tools this way, and this article is not a lab review of any single app. It is an explanation of a mechanism, plus a test you can run yourself in five minutes on any tool — including ours.

When is generative enhancement actually the right tool?

More often than its critics admit. The same "invents a plausible face" behavior is genuinely useful when fidelity was never on the table:

• Severe motion blur or tiny faces — a 40-pixel face in a group shot contains almost no recoverable identity; an invented-but-attractive face is strictly better than mush.
• Social sharing, not the family archive — for a profile picture or a quick post, "looks great" can legitimately outrank "is exactly true."
• Artistic or stylized output — if you want a polished, modern-looking portrait, a generative tool gets there faster than any faithful method.

The problem is not that these tools exist. It is that they are usually the default, and nobody warns you before you run your only photo of your grandfather through a pipeline that will quietly redesign his face. The right question is not "which app is best" but "which philosophy matches this photograph."

How can you tell if a tool will change a face before you pay?

You do not need to trust marketing copy — you can test fidelity yourself on one photo, at 100% zoom, in minutes. Any trustworthy tool makes this easy; a tool that hides the original is telling you something.

Two procedural checks matter as much as the pixel checks. First: does the tool show you the original side by side with the result, at full resolution, without a watermark blocking the comparison? A vendor confident in fidelity puts the evidence in front of you. Second: compare the result against a different photo of the same person, if you have one. Family resemblance survives across decades of photos; invented faces match nothing but themselves.

What should you use instead when the face actually matters?

When the photograph is the point — the only portrait of your father, the print going in the memorial frame — use a conservative restoration tool and judge it by the table above. The workflow that protects you: scan the original at 600 dpi first (never enhance a phone photo of a photo if you can avoid it — our digitizing guide covers the settings), keep the untouched scan forever, then restore a copy.

PhotoUnfade was built specifically for this case. It removes scratches, fading and yellowing while leaving facial structure, skin texture and age alone — the +29%/+77% sharpening calibration and the 0.0% color-invention result above are our pipeline, not a promise. We also keep restoration and colorization as separate steps, because a tool that "fixes and beautifies" in one click is a tool you cannot audit. And if blur rather than a changed face is your photo's main problem, our blur repair guide separates what sharpening can genuinely recover from what AI invents. If you want the broader landscape — which categories of tools exist, who each one is for, and the pricing traps to avoid — our honest buyer's guide lays it out, including five limitations of our own product. And if the photo is damaged enough that any software would be guessing, our AI versus professional comparison explains when a human retoucher is worth the money. For the full scan-to-finish sequence in one place, start with our step-by-step restoration guide.

The test costs nothing: three free restorations, no account, no credit card, full resolution, with the original always beside the result and a slider between them. Run your hardest photo — the face you know best — zoom to 100%, and decide with your own eyes. If we changed the face, you will see it immediately, and you should not pay us. That is the whole point of showing you both.