Are photo calorie apps accurate? What a photo cannot see

The doubt comes up in every forum thread about these apps: "it can't be accurate from a photo". The honest answer has two halves. A photo is accurate enough to be useful, and it is not a measurement. Knowing which half applies to the plate in front of you is the whole skill.

What the model actually sees

An image model is good at the first question, what is this, and much weaker at the second, how much of it is there. It recognises chicken, rice, avocado and sweetcorn in a bowl without trouble. It cannot see the oil the chicken was cooked in, the sugar in the sauce, how deep the bowl is or what sits under the top layer. Everything it returns is an estimate built on a reasonable guess about those invisible parts.

Where the error lives

Almost never in the identification. It lives in four places. Hidden fats: a tablespoon of olive oil is about 80 kcal and leaves no trace in a photo. Volume to weight: the camera sees area, not thickness, so a thick slice of tortilla and a thin one look alike. Mixed dishes: a stew's chorizo, a dressing on a salad, the cheese inside a sandwich. And density: 150 g of rice and 150 g of salad photograph as similar heaps and differ by hundreds of calories.

How big is the error, then?

We will not give you a percentage, because any single number would be dishonest: it depends on the plate. Grilled fish with boiled potatoes is an easy case. A plate of fried food swimming in oil, or a stew, is a hard one, and the same app can be close on the first and far off on the second. What an honest app can do is tell you how sure it is on each result. Fotocal labels every scan as an AI estimate with its confidence, 85 % on a clear chicken bowl for example, and it saves nothing until you confirm it. That label is not decoration: it is the signal to look twice.

Three corrections that remove most of it

First, the amount. If the estimate says 450 g and your plate is clearly a half portion, change it: the portion sheet multiplies the whole plate, ×0.5, ×1.5, or you type the grams, and calories and macros follow. Second, the ingredients. The result lists what it found with grams for each; remove the one that is not there, add the one it missed, fix the 150 g of rice that was 100. Third, the cooking method. Fried is not grilled: marking it changes the oil the app assigns to the dish. Those three taps fix more error than any model upgrade will.

For packaged food, do not use the photo at all. The barcode reads the manufacturer's label, which is the one figure nobody estimates. In Fotocal the barcode scanner is free and unlimited for that reason.

When a photo is the wrong tool

A sandwich you make every morning, a coffee, two eggs: say it or search it, do not photograph it. A home recipe you cook every week: weigh it once, save it, and reuse it. The photo earns its place at the restaurant, at a friend's table and in front of any plate you did not cook yourself, which is exactly where the other methods fail.

What to do with the number

Treat each meal as an estimate and the week as the measurement. A single scan being 80 kcal off is irrelevant; a week of scans trending 300 kcal above your target is information, and the scale will confirm it. Fotocal sets each scan against your own recent meals and your weekly average for that reason: the comparison is more robust than the single figure. Our guide to counting accurately covers the habits around the tool, and the food scanner page shows the real screens, including the correction steps.

So: are these apps accurate? Accurate enough to change what you eat, if you read the confidence, correct the obvious and judge by the week. The free plan includes two photo scans a day with no card, which is enough to find out on your own plates.