MyHeritage vs Remini: Which One Handles a Scanned Old Print

Two tools, two starting points

Most comparisons of MyHeritage vs Remini stack feature lists against each other, and the lists do not explain the result you end up holding. The two were built around different objects. MyHeritage grew out of genealogy, so its photo tools were shaped by family prints that will sit in a tree beside dates, names and places, often scanned from a shoebox at whatever quality the scanner managed. Remini grew out of mobile photography, tuned on faces from casual snapshots and selfies shot at arm's length, where the target is a crisp, flattering result viewed on a phone.

That origin shows up in the failure rate as much as in the wins. Feed each one a 600 DPI scan of a 1954 studio portrait and the difference is not subtle.

• MyHeritage tends to hold tonality across the whole frame, and it handles a scan where several people sit at different distances from the lens more gracefully.
• Remini tends to resolve a single face further, and it does best when that face fills a good part of the frame and was originally in focus.
• Both treat film grain as damage. That assumption is where most of the disappointment comes from.

Where MyHeritage handles an old print better

Three situations favour it. Group photographs with several faces, where attention has to spread evenly rather than concentrate on one. Prints with a colour cast or uneven fading, where large area tone correction matters more than edge detail. And any photograph headed into a family tree, because the output stays closer to what the scan actually recorded.

The last point carries more weight than it sounds. A restored portrait that sits beside a birth record is doing documentary work. A result that quietly sharpens an unfamiliar face into somebody the family does not recognise undermines the reason the photograph was scanned.

Where Remini wins on faces

Remini is the stronger choice when one face is the whole point, the original was in focus and reasonably close to the lens, and the result lives primarily on a screen. Modern snapshots that went soft, a school portrait that was slightly out of focus, a face cropped out of a larger group shot.

It also deals better with very low resolution input, where there is barely any facial structure left to work from. That said, it fills missing structure with invented structure more confidently than the other does, which brings us to the part worth learning.

The fingerprint test: what actually survived

Every print carries traces that identify it the way a fingerprint identifies a person. Its grain pattern, the weave of a jacket, the halftone dots in a newspaper clipping, the repeat in the wallpaper behind the subjects. None of this is decoration. It is physical evidence of how the image was made, and it is what separates a restored photograph from a plausible invention.

Enhancement models read grain as noise, because during training it was noise. Put the result next to the original at 100 percent and check three spots:

1. Cloth. Fabric should keep a directional weave. If it has become evenly bumpy, like orange peel, the weave was replaced with a texture guess.
2. A flat background. A wall, a sky, a studio backdrop. These should retain faint unevenness. Gone perfectly smooth means detail was generated rather than recovered.
3. The edge between subject and background. The transition stays slightly ragged in a real scan. A clean halo around every head means the outline was redrawn.

Run this test on both outputs with the original open beside them. Whichever result leaves more of those three intact is the one to keep, regardless of which looked better in the thumbnail.

Identifying people the honest way, using context rather than a sharpened guess, is covered in how to identify people in old photographs, and dating evidence behaves the same way in how to date an old photo.

Three things both get wrong

• Order of operations. Both apply sharpening early. Correcting exposure and tone first, then sharpening, keeps grain available as signal instead of destroying it.
• Colour on skin. Both push faces toward warm tones, which reads well on a phone and badly against a scan of a print that was never warm to begin with.
• A full frame treated as one problem. A group photograph has different needs per face. Running the whole frame through one setting means no part gets the right treatment.

The third one is fixable at home: cut the frame into crops, run each crop at its own settings, and reassemble. It takes longer and the result is noticeably better.

A workflow that uses both without paying twice

1. Scan first, at 600 DPI. Nothing downstream beats a better capture. This is the single step that decides the ceiling, and it is covered in how to scan old photos.
2. Correct tone yourself. Levels, white balance and a light dust pass. Do this before either tool sees the file.
3. Run MyHeritage on the full frame. Judge it against the original at 100 percent, not at thumbnail size.
4. Run Remini only on the one face that matters. Crop that face out, process it separately, then place it back.
5. Apply the fingerprint test to both. Keep whichever survived, even if the other looked better small.
6. Keep the untouched scan. Every derived version gets a new file name and the original never gets overwritten.

Steps two and six are the ones people skip, and they are the reason some archives are now stuck with only the enhanced version of a photograph nobody else has.

Our own earlier take on both competitors is in this versus MyHeritage in 2026 and this versus Remini in 2026. Where automatic work stops and hand work starts is argued in AI restoration against traditional work, archive context in genealogy and family history restoration, and keeping the result safe in heirloom photo restoration.

FAQ

• Which one is better for genealogy work? MyHeritage, mostly because it holds overall tone and does less single-face reconstruction. A photograph sitting next to a record should stay closer to what the scan captured.
• Is Remini safe to use on a one-of-a-kind print? Use it on a scan, never on the only copy, and keep the unprocessed file. It reconstructs confidently, which is exactly what you do not want applied to an irreplaceable original.
• Why do both make the skin look orange? They are tuned to warm faces because that reads well on a phone screen. Drop saturation slightly afterwards, or mask the skin and adjust it on its own.
• Can either one recover a face that is genuinely out of focus? Up to a point. Both rebuild plausible structure from training data. Past a certain blur there is no longer evidence to recover, and what comes back is a reconstruction.

← All articles