How accurate are AI photo calorie apps?

Published 21 Oct 2026 · source read 21 Oct 2026

In a 2026 NIH test, the photo features of four calorie apps underestimated the energy in meals by about 33%.

What the study measured

Researchers in the intramural program of the US National Institute of Diabetes and Digestive and Kidney Diseases (NIH) prepared 102 meals in a controlled metabolic kitchen, their content known from national food composition tables. They analysed photos of each meal with four apps — Appediet, Cal AI, Lose It! and MyFitnessPal — and compared every estimate with the known value.

What it found

Every app came in low. Mean energy underestimate per meal:

All four underestimated fat, by about 30 g; Lose It! and Cal AI also underestimated carbohydrate, by about 14 g. Lose It! and MyFitnessPal came closer on higher-energy meals than on lower-energy ones. The lead researcher's advice for a photo logged without checking portions: take the result "with a grain of salt".

What it does not show yet

The results were presented at the American Society for Nutrition's annual NUTRITION meeting as a conference abstract, not a peer-reviewed paper; the commentary in the report calls them preliminary until the full study is peer reviewed. The study measured accuracy, not whether using an app changes health outcomes.

What FatTonic does

FatTonic's camera names one fresh food at a time, and you tap the match to open its Food Index row, whose nutrients come from a named USDA or Fineli record. It does not estimate the calories of a mixed plate from a photo.