The answer depends entirely on what you are holding
There is no single accuracy figure for a painting identifier, and any app that quotes one is describing a benchmark rather than your situation. Performance splits sharply across three cases, and which case you are in is decided by the picture rather than the software.
A famous work, or a print of one, is named quickly and correctly. Those paintings have been photographed, catalogued and reproduced tens of thousands of times, so the match is essentially a lookup. A listed artist's typical subject is often placed to the right hand or at least the right circle, because enough published comparables exist to compare a style against.
An unsigned picture by a painter nobody recorded is the third case, and it is the one most people are actually in. There is no published record to match it to, and no amount of processing invents one. The honest output there is a period and a direction — nineteenth-century, Dutch school, probably a copy after a known composition — which is genuinely useful and is not a name.
Most pictures people ask about were never published, and an app cannot look up what was never written down.
Recognition is not attribution, and the gap is where the money is
This distinction does more work than any other in the field. Recognition says: this looks like Monet's handling, this subject, this decade. Attribution says: Monet painted this, and a valuation can be staked on it. The first can come from an image. The second comes from connoisseurship, materials analysis, documented provenance and a catalogue raisonné.
Authorship is the single largest lever on what a painting is worth — larger than size, condition, subject or medium combined. A canvas worth four hundred dollars as 'nineteenth-century French school' is worth a great deal more with a name attached and defensible, which is exactly why attribution is a slow, expensive, adversarial process rather than a photograph.
So the practical use of an app's answer is directional. If it points somewhere serious, the next step is a specialist in that artist — not a higher-resolution photograph and not a second app.
The two things a photograph cannot reach
The first is whether the picture is a painting at all. A good giclée on canvas photographs exactly like an oil, and this is the question that changes the figure most for the most people. An app will flag what it can see — a suspiciously flat surface, a visible plate edge, a dot pattern coarse enough to survive the camera — but a competent reproduction defeats it.
You can settle it yourself in two minutes with a ten-times loupe. A printed reproduction dissolves into a regular grid or rosette of dots under magnification; an original shows continuous, irregular brushwork, often with paint standing proud of the surface that catches in raking light. Then look at the edge of the canvas, where a print's image usually stops dead rather than wrapping.
The second is the back. Every experienced person turns a picture over first, because the verso carries the written history — gallery labels, exhibition stickers, framer's marks, stretcher stamps, inventory numbers, old auction lot tags. None of that is visible from the front, and it is what makes a name defensible rather than merely plausible.
- Under a 10× loupe: regular dots means print, continuous brushwork means paint.
- Raking light across the surface: raised paint and canvas texture, or nothing.
- The canvas edge: an original's paint usually wraps, a print's image stops.
- The verso: labels, stamps and numbers, photographed close and in focus.
- The frame, which dates pictures surprisingly well.
Where the instrument is genuinely good
It is worth being specific about the upside, because the sceptical framing can obscure it. On a museum wall with an unreadable label, on a phone, in ten seconds, an identifier is excellent — and that is a real situation that a great many people are in a great many times.
At an estate sale with four minutes to decide, it converts an object you cannot name into a search you can run. That is worth more than it sounds: the hard part of researching an unknown picture is not the research, it is knowing which three words to type first.
And on the third case — the attic picture — an honest instrument that says 'period and school, not a name' has told you something true and saved you from the confident wrong answer that a search engine's image match will happily give you instead.
How to photograph a painting so the reading is worth having
Most poor results are poor captures. Varnish, glare and angle destroy more readings than the subject matter does, and all three are free to fix.
- Square on to the surface, no angle, so the composition is not distorted.
- No flash. Varnish throws it straight back and erases a third of the picture.
- Indirect daylight, and one extra frame raking across the surface for texture.
- Close frames of the signature and any monogram, in focus.
- The whole reverse, including labels and stamps.
- Take the glass off if it is safe to do so — glazing reflects everything.
What to do with the answer
Treat it as a hypothesis with a confidence attached. If the app names a catalogued work and the picture is a plausible size and medium for it, you very likely have a reproduction of something famous, and that is worth knowing before you get excited.
If it points at a listed artist, look up recent auction results for that artist at that size and period — results, not asking prices, which are not evidence. Then photograph the front, the back and the signature and send them to an auction house specialist, who will look at photographs without charge because they are hoping to sell it.
If it gives you a period and no name, that is the true answer and the app has behaved correctly. The next step is a local specialist or a regional auction house, not more software.