OldPhoto vs VanceAI: Two Different Ways of Repairing an Old Photo

What VanceAI actually is

VanceAI is built as a set of separate modules. Upscaling, denoise, sharpening, old photo repair, background removal and a few others each sit in their own box, each drawing credits per image, or all of them covered by a subscription. That shape suits occasional use well, because you are not paying for modules you never open.

For restoration it has one practical consequence. The repair module handles repair, and enlargement means running a second module and spending again. A print that needs both costs two runs and two exports, and on old photographs those two steps usually travel together. The order is set out in how to upscale old photos.

Credit balances also carry a quiet cost. Whether you repair twenty prints in one sitting or one print on twenty separate days, the per-image model charges the same, but the experience is nothing alike.

What OldPhoto does instead

One input, a damaged photograph. One output, a repaired one. No module to choose. The model has seen a great many damaged prints paired with undamaged ones, and it infers a plausible clean image, which makes it strong on statistical problems: dust, light scratches, mild fade, flat contrast, slight softness. Those are the failures that show up most often.

Where it is weak is where only one answer is right. Half a missing face gets filled with something statistically average, and a uniform gets a plausible colour that may belong to the wrong regiment. That is the same failure mode written up for colour in how accurate colourisation is, and it is cheaper to accept before starting than to discover afterwards.

Stacked passes are the second practical limit. Enhancement run twice rarely improves anything, because skin goes waxy and fine detail turns plastic. One clean pass from a good scan beats three stacked attempts.

Compared by damage type

• Surface dust and light scratches: both fine, and the difference is not worth choosing over.
• Fading and flat contrast: both fine. Pulling curves is mechanical work either way.
• Repair plus enlargement: VanceAI bills two modules, OldPhoto finishes in one flow. This combination is common on old prints.
• Water damage and lifted emulsion: neither is enough. Filling a void needs a judgement about what belonged there. That damage is covered in water damaged prints.
• A badly damaged face: neither is enough, and this is the case that sends people to a professional. The cost side is in free versus paid.
• A few dozen prints at once: the per-image model gets expensive in a linear way, and the batch route is cheaper by a wide margin. Triage thinking is in album digitisation.
• Needing to explain every change later: neither tool gives you that. Layers and history are the evidence, and a model pass is not, which is set out in AI versus traditional restoration.

The bottom four rows of that list matter more than the top three. The top rows are work both tools do, and either is fine. The bottom rows decide whether you finish in ten minutes or go looking for a person.

How the money works

The shape of it: VanceAI charges per image against a credit balance, or a subscription covering every module. OldPhoto charges by volume, with no second charge for chaining steps. Exact figures move with region and plan, so read the following as shape rather than quote.

• Three or five prints a year, and you also want upscaling and denoise: the suite is the better buy, since one subscription covers everything.
• A shoebox of two hundred that only need repairing: per-image billing gets expensive here, and the batch route is clearly cheaper.
• Both: the common arrangement. Run everything through the cheap pass, then pay a person for the few you intend to frame. The cost side sits in free versus paid.

Getting the best out of either

Order matters more than the tool. Follow these six steps and most failures show up before they cost you anything.

1. Scan first, and scan properly. Nothing downstream recovers detail the capture never had. Clean the glass and scan higher than you think you need. The details are in how to scan old photos.
2. Run the first pass on a copy. Never the original file.
3. Look at it at 100 percent before doing anything else. Eyes, mouth line and hairline are where invented detail shows up first.
4. If it needs enlarging, do that after repair, not before.
5. Send the cleaned file to a person if it still needs work. A retoucher starting from something cleaned costs less than one starting from something filthy.
6. Keep the versions separate. Original scan, automated pass, hand finished file, clearly named. A scheme that survives contact with a real family is in archive organisation.

How to decide about one specific print

Ask what the image has to do. A photograph going into a family history with dates and names is a record, and invented detail in it becomes somebody else's problem later. A photograph being reprinted for a birthday is a keepsake, and plausible is plenty. The same split, from the tooling side, is in app versus desktop workflows.

FAQ

• Does the VanceAI repair module cover everything, or do I need other modules? Repair plus enlargement covers most old prints. Denoise and sharpening depend on the individual scan rather than being needed every time. Check at 100 percent before adding modules instead of running all of them up front.
• Can OldPhoto replace the whole VanceAI suite? No, and it does not try to. Upscaling for ordinary images, denoise and background removal are not part of restoring old photographs. If you also handle general image work, the two can sit side by side.
• Is trying both on the same file a bad idea? On the same original, yes. Repeated compression and repeated enhancement each remove a layer of fine detail. Run each on a copy, compare, then continue with one result rather than stacking them.
• Should the finished restoration be labelled? Yes. Anything colourised or heavily inferred should be labelled as an interpretation when published, and any print going into a family history should keep the original scan as the record. The wider trade-off is in AI versus traditional restoration.

← All articles