There is no accuracy percentage that fits every meal
An AI calorie counter can recognize visible foods and suggest quantities, but a photo does not reveal a recipe or directly measure calories. A useful estimate depends on three things: identifying the right foods, estimating how much you ate, and using nutrition values that match those foods.
“Accurate” can mean different things. A system can correctly identify rice while misjudging its weight. Another can get a meal total close by overestimating one ingredient and underestimating another. Neither tells you that every item or macro is correct.
A 2024 comparative study of nutrition apps evaluated food recognition and nutrient estimates separately. It did not test Heyberry, and its results should not be treated as an accuracy score for this app or for every meal you might eat.
Why a clear photo can still produce a wrong estimate
Portion depth is hard to see. Two bowls with the same width can hold different amounts. Food stacked behind another item is partly hidden. A photo with no reliable scale reference can make the amount harder to judge.
Ingredients can be invisible. Oil used in cooking, butter inside a sandwich, sweetened drinks and dressing mixed through a salad may not be identifiable from the surface. Describe what you know rather than expecting the camera to recover it.
Similar foods have different nutrition. Plain yogurt and sweetened yogurt, skinless chicken and chicken with skin, or a light sauce and a cream sauce can look alike. Brand and recipe information matter when they are available.
Raw and cooked weights are different. Cooking changes water content. A weight entered against a raw-food record may not describe the same amount of cooked food. Check which form a nutrition record refers to before comparing totals.
Review these four things before saving
- Food names: does each item match what you ate? Look for missing sides, toppings, sauces and drinks.
- Amounts: compare the suggested grams with a known serving, package quantity or weighed portion. A familiar food name does not validate the amount.
- Preparation: check frying, added fats, skin, breading and other details that affect the meal.
- Coverage: include everything you ate, but avoid logging the same ingredient twice through both the recipe and a separate addition.
If a result seems surprising, start with the ingredient breakdown. Changing the total alone hides the cause and makes it harder to reuse the meal consistently.
When to use a barcode or recipe instead
For a packaged snack, the label and the amount eaten are usually more informative than a photo of the snack. Scan the barcode, then confirm the product and serving with the package. Databases can be incomplete or use older labels.
For food you cooked yourself, known ingredient quantities provide a stronger basis than estimating a mixed dish from its appearance. Add the ingredients for the whole batch and divide by the portion you ate. See how to count calories in homemade food for a worked example.
At a restaurant, exact ingredients and amounts may be unavailable. Keep the result as an estimate, use any published restaurant nutrition information, and add preparation details you know. Changing camera angles cannot reveal an unknown recipe.
What Heyberry lets you review
Heyberry shows the identified foods, estimated grams, calories and macros. You can change an item's grams, answer portion questions, or use Fix with a note to describe a correction such as “no dressing.” Review the new result after a correction; reanalysis is still an estimate.
We have not published an independent accuracy benchmark for Heyberry. We do not assign it a fixed accuracy percentage or promise that it is more accurate than another app. The photo calorie counter page explains the logging flow and Pro pricing.
A useful first test
Choose a simple meal with ingredients and portions you already know. Record those details, take a clear photo, and compare each predicted item with your reference. Repeat with a mixed meal before deciding whether this method suits your routine. One good result is not a benchmark.
These tools are for general wellness logging. A photo estimate is not a basis for medical dosing or a substitute for nutrition care tailored to a medical condition.