01 — Photo logging

One photo. Whole meal logged.

CalorieCounter is an AI calorie counter for iPhone. Point the camera at your plate and it names the foods, guesses the portions, and tells you how sure it is about each one. Every number stays yours to change.

By Alen Yaco · Double bachelor's in biological sciences and social ecology, UC Irvine · Published 17 September 2026

What happens when you scan a meal?

You open CalorieCounter, tap the scan button, and fit the plate inside the corners. About 4 seconds later the app lists each food it can see, an estimated portion in grams, and the calories and macros that follow. Every row is editable before you save it.

1. Frame the plate. Open the app, tap the scan button, and fit the whole plate inside the corners. Good light helps; a perfect angle does not matter.
2. The photo leaves your phone clean. Location and device metadata are removed on the device before the image is sent for analysis.
3. A result in about four seconds. The vision model returns each food it can see, an estimated portion in grams, and the calories and macros that follow from it.
4. Every item wears a label. Pretty sure, Best guess, or Check this. The label tells you where to look before you save.
5. Fix, then save. Nudge a portion, swap an item, or move the meal to another slot. Save, and the Today ring updates. The correction is remembered for that food next time.

How accurate is an AI calorie estimate?

Accurate enough to be worth saving, and honest about where it is not. CalorieCounter never publishes a single accuracy percentage, because one number hides the meals it gets wrong. Instead every item on a result carries one of 3 labels: Pretty sure, Best guess, or Check this.

Here is what each label means:

A wrong number should look wrong, not authoritative. That is the whole idea.

What does photo logging handle well, and what does it miss?

Photo logging is strongest on plates with separate, visible components, and on restaurant food that is in no database. It is weakest on mixed dishes, on oils and dressings that add calories without adding much to see, and on anything hidden under other food.

Where photo logging shines

• Plates with visible, separate components: protein, starch, vegetables.
• Everyday meals you eat often, because corrections are remembered.
• Restaurant plates you would never find in a database.
• Fruit, snacks, and anything with an obvious shape and size.

Where you should check the label

• Mixed dishes such as soups, stews, and casseroles, where the model estimates the whole rather than each ingredient.
• Oils, dressings, and sauces, which add calories without adding much to see. Peel will remind you: the sauce counts.
• Portions hidden by other food, or served in deep bowls.
• Packaged food. Use the barcode instead; it is exact and free.

Can you correct what the AI gets wrong?

Yes, and the correction sticks. Every item on a result is a row you can edit in place: portions move in 5 g steps or you type a number, and Match to a food swaps a wrong item for a catalog entry. Save, and the app remembers that fix for that food.

If an item belongs to another meal, move it there. Corrections are remembered per food, so a plate you eat often scans closer each time.

How many AI photo scans do you get for free?

The free plan includes 2 AI photo scans a day, which reset at midnight in your own time zone. Barcode scanning, food search, quick add, and voice or text logging stay unlimited on every plan. Premium raises photo scans to 60 a day and re-examines uncertain results.

That second look re-examines any result labeled Best guess or Check this. See what stays free.

What happens to your meal photos?

Location and device metadata are stripped on your phone before the image leaves it. Photos go to Google Gemini through our servers, are kept for 30 days so a Premium re-analysis can look at the same image, then deleted. They never train Google’s models or ours.

Photo analysis can be switched off in Settings while every other way of logging keeps working. The details are on the privacy page.

Sources

• Google Gemini, the vision model that estimates photographed meals
• Open Food Facts, the open database behind packaged products
• USDA FoodData Central, the reference database behind generic foods