How to choose an antique identification app

There is no single best antique identification app, because identification is really four separate jobs: naming an object, reading a mark, finding what comparable things have sold for, and getting a written opinion from a person who will put their name to it. Decide which job you are in, and the shortlist usually collapses to one or two tools in about a minute.

This is why roundups of the top ten antique apps tend to leave you no wiser. They rank a reverse image search against a sold price archive against an online appraisal service, as though those were competing answers to the same question. They are not. A price archive cannot tell you what you are holding, and an identification tool cannot responsibly tell you what it is worth.

So start with the moment you are in, because the moment dictates the tool.

• You are crouched in a stranger's dining room with a lidded jar in your hands, four minutes before someone else picks it up, and the only question that matters is whether this is actually antique or just old.
• You are at your own kitchen table with something inherited, no clock running, a mark you can half read, and an afternoon to spend on it.
• You already know what it is and you want market context before you list it, insure it or divide it between family.
• You need a document, because an executor, an insurer or a sibling is going to read it and act on it.

Those four situations do not share a best app. A tool that is excellent in the dining room is the wrong tool for the document, and the reverse is just as true. The rest of this guide works through them in that order, then gives you a way to test any candidate yourself in five minutes.

Is this worth anything is the question almost everyone actually types, and it is two questions wearing one coat. The first is identification: what is this thing, who made it, roughly when, and where. The second is valuation: what would someone pay for this example, in this condition, in this market, this month.

The first question can be worked on reasonably well from good photographs, because a lot of the evidence is visible. Form, proportion, material, glaze, joinery, casting seams, the pattern of wear and the mark itself are all things a camera can record. The second question is a market judgement about one specific object, and it turns on things a photograph does not carry: condition under raking light, restoration, completeness, provenance, and what similar pieces have been doing at sale recently.

That is why a figure produced from a snapshot is the least trustworthy output any tool can hand you, and why we think an identification tool should refuse to produce one at all. A number invites you to act. A name invites you to look closer, which is the correct response to an object you have known for ninety seconds.

So when you compare apps, notice which of the two questions each one is answering, and be wary of anything that answers both with the same air of certainty. Identification is one discipline. Valuation is another, and it belongs to appraisers who will examine the object and sign the opinion.

Ask which antique app is most accurate and you will get a lot of confident answers and very little evidence. There is no shared, public test set of photographed antiques with verified attributions, so any single accuracy percentage you meet should be read as marketing rather than measurement. We are not going to publish one either, because we would be making it up.

What you can evaluate, in five minutes, is how a tool behaves at the edges. That turns out to matter more than any headline figure, because there are two ways an identification fails and they hurt in different ways.

Overconfidence

The tool hands you a maker and a date with no caveat, you believe it, and you either pay too much for a reproduction or pass on something good because the answer sounded settled. This is the expensive failure, and a visible confidence level exists to prevent exactly it.

Useless hedging

The tool tells you it is a ceramic vessel, possibly European, possibly twentieth century. Technically defensible, and entirely worthless to someone deciding whether to hand over a note in the next three minutes.

The useful middle is a specific reading with its uncertainty attached, and a clear signal about what to do next. That is the standard we hold ourselves to, and the one we would suggest you hold any tool to: a second opinion rather than a verdict, precise about what can be seen in the photograph, and plain about what cannot.

If you want the longer version of this argument, the companion page on how accurate AI antique identification is goes through what a confidence level can and cannot tell you, and where these models tend to fail.

Two people can photograph the same creamer and get usefully different results, and most of that difference is the photograph rather than the software. An identification tool reads what the image contains, so the quickest way to improve any of them is to give it more to read.

• Shoot the base and the mark separately. A mark that occupies forty pixels in a full object shot carries almost no information. Fill the frame with it.
• Light it from the side. Flat overhead light erases the shallow relief that tells an impressed mark from a printed one, and hides the tool marks and wear that date a piece.
• Include a straight on profile. Silhouette and proportion carry period. A three quarter view flatters the object and confuses the outline.
• Photograph the construction, not just the surface. The back of a frame, the inside of a drawer, a joint, a seam, a foot rim. This is where age is honest.
• Keep the background plain. A patterned tablecloth competes with the pattern you want read.

None of this is unique to one app, which is why it belongs in a guide about choosing one. If a tool gives you a disappointing result, retake the photograph before you conclude the tool is poor, and retake it before you conclude the object is ordinary. Our page on scan capture covers how to get a usable photograph when you are standing in a crowded room with one hand free.

Free tiers in this category almost always meter volume rather than capability, and that is the right way round. You should be able to test the thing that matters, the quality and honesty of a result, before you pay anyone anything.

So when you compare pricing, ask two questions. Does the free tier include the full identification, or a teaser of it. And does the paid tier add anything other than more scans. If a paid tier is the only way to see a confidence level or a disclaimer, that is a product decision worth noticing.

Ours works like this, and we would rather you knew before you signed up than after. Patina's Free plan costs nothing and allows three scans per calendar month, enforced on the server, and includes maker, era and origin from a photo, an honest confidence level on every result, an appraiser referral when a result needs an expert eye, and a scan history of your last ten scans. There is no capability behind the paywall, only volume.

That shape follows the two readers of this guide. Three a month suits the occasional find, the inherited box, the thing you have been wondering about since spring. Unlimited suits the estate sale regular who wants to scan every unfamiliar object they meet on a Saturday morning without counting.

No app in this category replaces an appraiser, and the better question is when the cost of one is proportionate. That depends on the object, and it depends on what you need the answer for.

Fees vary by appraiser and by the kind of report you need. One appraisal guide puts an initial appraisal by a certified appraiser at roughly $50 to $200, and a comprehensive appraisal at $300 to $1,000 or more, and notes the final cost depends on the appraiser's experience and the services you ask for, which is why it recommends getting more than one quote. Treat those as orders of magnitude rather than a price list, and get quotes for your own object.

Against that, the arithmetic is usually clear. For a small curiosity, a formal appraisal costs more than the object and you are better served by identification plus your own reading. For something that may be significant, for an estate being divided, for an insurance schedule or for anything where another person has to rely on the answer, an identification app is a triage step and the appraisal is the deliverable.

That is the logic behind a triage flag rather than a confident guess. The useful output when a tool cannot place an object is not a vaguer answer, it is a clear statement that this one needs an expert eye and a route to one. When Patina shows a referral, the link goes to an independent service we have no partnership with and carries a tag naming Patina, never anything about you or the object you scanned.

Two definitional points cause more confusion in this category than any software limitation, and knowing them makes every result you read more useful.

The first is what counts as an antique at all. The customary definition is an object at least a hundred years old and in original condition. Wikipedia's article on antiques notes that the 1930 Smoot Hawley Tariff Act exempted works of art and objects produced before 1830, a cutoff that at the time of enactment amounted to roughly a century of age. Vintage, by contrast, has no equivalent settled line. So when a result tells you a piece is mid twentieth century, it has answered your question even though the word antique never appeared.

The second is that a mark dates a pattern or a factory, not necessarily your example. Marks were used for long stretches, reused after takeovers, copied outright, and applied to later production of an old line. A legible mark narrows the field considerably and almost never closes it on its own. This is why the form, the material and the construction matter alongside the mark, and why a tool that reads only the mark is working with one witness.

Both points lead to the same practical habit. Treat an identification as a hypothesis with a confidence attached, then go looking for the second piece of evidence that would confirm or break it: the joint, the foot rim, the seam, the weight in the hand. That habit costs nothing and it is what separates a collector from a buyer.

If you have decided which job you are in, the rest of this pillar goes deeper on each one, and those pages do not repeat this guide.

• If you want to see what a result looks like, the page on photographing an object and getting maker, era and origin in plain English walks through a single identification end to end.
• If your worry is the photograph, the scan capture page covers taking one usable shot where you are standing.
• If your worry is reliability, the page on how accurate AI antique identification is handles confidence, limits and the honest failure modes.
• If you came here from a general image search, the Google Lens page sets out the comparison between visual search and domain specific identification in detail.
• If you have a house or a box to work through, the page on identifying inherited antiques one box at a time is about sequence and sanity rather than single objects.
• If you want to keep what you find, the scan history page explains how your last ten scans sit in your account.
• If you want the data answer first, the page on what happens to the photographs you scan covers storage, retention and deletion.

And if you want the one sentence version of this whole guide: choose the tool that matches the question you are actually asking, and prefer the one that tells you when it is unsure.