PlantNet or PlantSnap: free research against a paid reference set

PlantNet asks a question before it identifies: which organ did you photograph — leaf, flower, fruit, bark, whole plant, habit? PlantSnap does not.

That is not friction for its own sake. A leaf photograph and a flower photograph carry different information and are compared against different reference sets, so telling the app which it has narrows the candidate space enormously. It is exactly what a botanical key does at its first branch.

Skipping the prompt is the commonest reason people try PlantNet once and conclude it is worse. Answer it, and the result improves visibly.

PlantNet returns a list with similarity percentages. PlantSnap returns a species.

The list is more useful whenever being wrong has a cost. Identification errors are not random — they cluster on look-alike pairs — and seeing that the second candidate is 41% where the first is 47% tells you the model is guessing between two things, which is exactly when to check a diagnostic character yourself.

A single confident name conceals that. It is a better experience and worse information, and which you want depends on whether the plant is a specimen or a pot.

PlantSnap was built when separating a subject from a cluttered background was the hard part of computer vision, and it still prefers a single leaf or flower filling the frame against a plain background. PlantNet is more tolerant, partly because the organ prompt is doing work the framing would otherwise have to do.

The discipline improves both, and it costs nothing: one subject filling the frame, even light, no flash, and the most informative organ first — flower, then leaf, then fruit, then bark.

The habit worth building is taking three photographs rather than one: leaf top, leaf underside, and the twig showing how leaves attach. Arrangement is the first branch of every real botanical key and neither app asks for it, which is why our free leaf key does.

On 2026-09-03, PlantNet listed no in-app purchases at all and PlantSnap listed 10 tiers between $0.99 and $59.99.

PlantNet has no paid tier because it is not a business: it comes out of public agricultural and development research, its purpose is a shared scientific database, and observations you contribute feed it. That also means your identifications have a second life as data.

PlantSnap is a commercial product and the subscription is the product. Apple does not publish billing periods next to in-app purchase names, so read the tier list below as a shape — several tiers sharing a name is one product billed weekly, monthly and annually, and the lowest number is the shortest term.

Neither covers cultivars well. PlantNet is built on wild flora by design; PlantSnap is broader but still not a horticultural database. If the plant came from a garden centre and you need the named variety, PictureThis is the app with that reference set.

Neither works offline — both identify server-side. Seek is the free app that identifies on the device with nothing uploaded, and it covers everything alive rather than only plants.

And neither will confirm an identification with a human. iNaturalist, also free with no in-app purchases, gets naturalists to check it, which is the most reliable answer available from any app.

Prices, tiers, ratings, versions, sizes and update dates are read from each app's US App Store listing on 2026-09-03 through Apple's Lookup API and product pages. Nothing is typed from memory and the build fails if a price appears in the prose.

The two apps do different things with your data, and it is worth knowing which you are taking part in.

PlantNet is a citizen-science project. Observations you choose to share go into a shared scientific database used in botanical research, with your identification weighted against other contributors' — the same model iNaturalist uses. Contributed data has fed published work on species distribution and phenology, and the reference set that will identify your next photograph was built the same way.

PlantSnap uses submissions to improve its commercial model, which is standard, disclosed, and produces no public dataset.

Neither is wrong and the day-to-day difference is small. But if the idea of a photograph having a second life as data appeals — or if it distinctly does not — that is a real basis for choosing, and it never appears in a feature comparison. If you would rather nothing left the phone at all, Seek identifies on the device and uploads nothing.