Where Mira gets her calorie numbers

Mira first looks for the food in your diary and in the Calorize catalog. If nothing similar is there, she estimates it from general knowledge of that kind of food — commonly accepted average values — and creates a new product. That isn't lab precision, and it doesn't need to be: errors in both directions largely cancel each other out over a week.

Where the numbers come from, in order

For every dish or drink, Mira goes down this list and stops at the first thing that fits:

  1. Your diary. Something already in your diary, at the same portion as last time. Saying "coffee" is enough if your coffee is already there.
  2. The Calorize catalog. A shared base of foods and dishes, store products from labels, and your own products and recipes. Things that have already appeared in your diary come first in search.
  3. A label or barcode. Photograph the nutrition table or scan the barcode, and the manufacturer's data is used.
  4. Mira's general knowledge. When the catalog has nothing similar — more on this below.

Mira always tells you where a product came from: whether she found it in the catalog or created a new one.

What "general knowledge" means

Mira runs on a large language model, like other AI assistants. The model was trained on a huge amount of text, including nutrition tables, recipes, labels and cooking references. So it knows what a burger, a cappuccino, a potato salad or a pilaf is usually made of, and roughly how many calories, protein, fat and carbs that food has on average.

When a specific dish isn't in the catalog — a burger from a particular café, a homemade pie, an unusual drink — Mira:

  1. Takes commonly accepted average values for similar food, per 100 g.
  2. Estimates the portion: from what you said ("200 g", "a slice", "a bowl"), or from a usual serving — a bowl of soup ≈ 300 g, a banana ≈ 120 g, a cup of coffee ≈ 200 ml.
  3. Creates a new product and logs it. Next time, that product is in your history and Mira uses it.

Example: "a burger at a café". That exact burger isn't in the catalog, so Mira takes a typical burger — about 250 kcal per 100 g, a portion of around 220 g — and logs ≈ 550 kcal.

A photo of a dish works the same way: the recognition module estimates what's on the plate and how much it weighs, and takes average values for each component. In our own checks — the same dish both weighed and photographed — the photo estimate usually comes out a little higher than reality. That's the better side to err on: under-counted food hurts your goal more than slightly over-counted food.

How accurate is this — and why that's fine

Food calories are never perfectly precise — not even where they look precise:

  • Labels are approximate too. In many countries manufacturers are allowed to be off from the real composition by up to roughly 20%.
  • The same dish varies. A borscht from two kitchens, a portion from two cafés, a spoon of oil more or less — the difference easily reaches a hundred calories.
  • An eyeballed portion is an estimate. Even with kitchen scales, nobody weighs every spoonful.

And that's fine, because the errors pile on top of each other and largely cancel out. Sometimes Mira estimates a bit more than there really was, sometimes a bit less — and over a week it evens out. What matters for results isn't the precision of each entry but consistency: a rough entry every day is more useful than a perfect one once a week.

The best judge is the scale. If you log food in roughly the same way, within 2–3 weeks you'll see whether your weight is moving as planned. If not, a small adjustment to your goal is enough — how to do that is in Goals & Calculations.

So there's no need to count down to the last calorie. Sometimes all you have for a whole day is a rough number — you can just tell Mira that too: "today was about 2500".

Describing food when there's no scale

A kitchen scale is the most accurate — nothing better has been invented. When there's none at hand, a good description beats a photo. Three things matter most:

  1. What exactly. Not "chicken" but "chicken thigh" or "chicken breast"; not "porridge" but "oatmeal made with milk".
  2. How fatty or lean, how sweet. This is where most of the calories hide. Is there a film of grease on the soup, or is it clearly lean? Is the shake very sweet or barely sweet?
  3. How much. No grams needed: a big, medium or small plate; a thick or thin slice of bread; a handful of salad; a piece the size of a matchbox.

A photo of a dish is for when you can't do better — say, in a restaurant. Mira estimates from the photo, but she can't see the oil in the pan or the sugar in the sauce, so expect a rough figure. At home, weigh it or describe it.

How to make it more precise when you want to

  • Say the weight. "200 g" is more precise than "a bowl".
  • Say what exactly. "Cappuccino with oat milk", "burger with two patties and bacon", a brand or a place.
  • Photograph the label or scan the barcode — then the manufacturer's data is used.
  • Give exact numbers if you know them. For example: "protein bar, 200 kcal, 20 g protein" — Mira will create the product with exactly those values.
  • A homemade dish with known weights is a recipe. When to create a product and when a recipe is in Product or recipe.
  • If Mira picked the wrong thing — just say what's off and she'll fix the entry.

Didn't find your answer? Drop us a message — every request gets a reply.

Other topics

We use our own cookie and analytics tools to understand how people find and use Calorize. If you came from a Google ad, we tell Google when you sign up or buy, to measure that ad. There are no ads inside the app. Your choice is remembered for a year.