Photo check

Recognise any fruit from a photo

Upload a clear photo and get a species and variety read, a confidence level, the features the model used, and guidance on what would sharpen the answer.

Choose a photo

Drag a photo here, or tap to choose. A clear, close shot of the whole item works best.

Fruit recognition apps feel like magic the first time they work. You point your phone at something unfamiliar, and a name appears. The second time you use one, you notice the limits — it confidently names the wrong variety, or calls a quince a pear, or gives up entirely on a fruit you have cut in half.

Understanding how these tools actually work makes them far more useful, because you learn when to trust the answer and when to double-check it.

How Recognition Actually Works

Every fruit recognition app is doing the same underlying job: converting your photo into a set of numerical features and comparing them against patterns learned from many thousands of labelled images.

The model is not reasoning about botany. It has not learned that a mango has a single large flat seed and belongs to the cashew family. It has learned that certain arrangements of colour, texture, shape and edge are statistically associated with the label "mango". That distinction explains almost every failure mode you will encounter.

Where It Is Reliable

  • Common cultivated fruit, whole, in good light. Apples, bananas, oranges, mangoes, avocados, strawberries. Accuracy here is high enough to be genuinely useful.
  • Distinctive shapes. A pineapple, a starfruit, a dragon fruit, a durian — these are hard to confuse with anything else.
  • Telling categories apart. Citrus from stone fruit, berries from drupes, tropical from temperate.

Where It Fails

Variety level. This is the biggest gap. A model can reliably say "apple". Saying "Pink Lady" rather than "Fuji" requires reading skin finish, blush pattern, shape and size together — features that overlap heavily between varieties and vary with growing conditions. Our apple variety guide sets out why this is genuinely difficult rather than just a matter of a bigger model.

Discover more
Get Produce Boxes
fresh produce
produce

Unripe fruit. Immature fruit often does not look like its own species yet. Green tomatoes, unripe mangoes and young stone fruit all cause trouble, because the reference images are overwhelmingly of ripe fruit.

Cut, cooked or processed fruit. Once you remove the skin, the features the model relies on disappear. A sliced apple and a sliced pear are close to identical in a photo.

Lookalikes. Quinces and pears, lemons and limes, tangerines and mandarins, sweet and sour cherries. These are visually close enough that a photo alone is often insufficient.

How to Take a Photo That Works

  • Fill the frame. The fruit should take up most of the image. A fruit in the middle of a wide shot gives the model very little to work with.
  • Use even, natural light. Direct sun creates blown highlights and deep shadows that hide texture. Open shade is ideal.
  • Plain background. A cluttered bench adds visual noise the model has to ignore.
  • Include a size reference. A hand or a coin gives scale, which separates small citrus from large ones.
  • Take two or three angles. Stem end and side view together carry far more information than either alone.
  • Photograph it whole if you can. Skin carries most of the identifying information.

Why Confidence Scores Matter More Than Names

A recognition app that always gives an answer is worse than one that sometimes declines. The always-answering app produces false confidence, and false confidence about food is a genuine problem — particularly for anyone foraging, or managing an allergy.

Look for a tool that gives you a confidence figure and explains which visual features it used. That explanation is also how you learn: after a dozen uses you will start noticing the same features yourself, which is the point. Our guide to produce identifierscovers how to read those outputs.

Beyond Recognition

Naming a fruit is the beginning of a useful answer, not the end. The follow-up questions — is it ripe, how long will it keep, what is it good for — are where an app earns its place. Ripeness in particular is something a recognition model can help with, because many of the ripeness cues are the same visual features used for identification.

For fruit that is already in the kitchen, the most valuable thing an app can do is help you use it. If you want that, see how a fridge photo turns into recipes rather than reaching for another identifier.

The Short Version

Fruit recognition works by matching visual patterns, not by reasoning about botany. It is excellent for common whole fruit in good light, weak at variety level, and unreliable for unripe, cut or cooked fruit. Take a close, well-lit photo against a plain background, include a size reference, and trust a tool that admits uncertainty over one that never does.

Frequently Asked Questions

For common whole cultivated fruit in good light, accuracy is high. For variety-level identification, unripe fruit, or anything cut and cooked, accuracy drops sharply because the visual features the model relies on are missing or ambiguous.