AI calorie counter: how a photo of your plate becomes an estimate

How does an AI calorie counter work?

Almost every app that counts calories from a photo follows the same five steps. Knowing them shows you where an estimate is solid and where it needs your help.

1. The photo. You take a picture of your plate. The model only knows what the camera sees, so light, angle and framing matter more than you might think.
2. The items. An image model splits the plate into foods it recognises: “chicken shish”, “rice pilaf”, “shepherd's salad”. A dish that's eaten as one thing, like a stuffed pepper, stays one item.
3. The portions. For each item it estimates an amount in grams, judging size against the plate, the cutlery and what a typical serving looks like. This is the hardest step, because a photo is flat and food has depth.
4. The nutrients. Grams become calories, protein, carbs and fat from what the model knows about that food per 100 g, much like reading a calorie chart.
5. Your review. You check the list, fix a name, nudge the grams or remove something that isn't yours, then save. This step turns a guess into a log you can trust.

So an AI calorie counter never weighs anything. It recognises, estimates and adds up, and it asks far less of you than weighing and searching every food, which is the whole point: a log you'll actually keep. The other half of the picture is your daily target, which the calorie calculator estimates.

What does a photo calorie estimate look like?

Here is a typical kebab plate, worked out the way a photo estimate does it. The values per 100 g come from our calorie chart: chicken shish kebab 150 kcal, rice pilaf 145 kcal and shepherd's salad 55 kcal.

The chicken is the easy part: clear shape, clear size. The pilaf and the salad are less certain, because no camera can see how much butter went into the rice or how much olive oil is waiting at the bottom of the bowl. That's exactly where your review earns its keep, and why we wrote a whole guide on how accurate AI calorie counters are.

What makes a good photo for counting calories?

You don't need a studio. A few small habits change a lot about what the model can read from your plate:

• Shoot before you eat. A full plate is far easier to read than a half-eaten one. If you leave some, lower the grams afterwards.
• Get the whole plate in. The rim helps the model judge size. Then crop away the table, your phone and your friend's dessert.
• Use soft, even light. Daylight or a bright room beats a dim restaurant corner. Avoid hard shadows across the food.
• Angle it a little. Straight down shows the layout; a gentle angle shows height. For deep bowls of soup or stew, the angled shot tells more.
• Leave a fork or spoon in the frame. Everyone knows roughly how big a fork is, the model included. It works like a ruler.
• One plate per photo. Photograph what you'll actually eat, not the whole shared spread.
• Say what the camera can't see. “Cooked in butter”, “two sugars in the tea”, “cheese inside”. A short hint fills the gaps a picture can't.

When is a barcode or typing better than a photo?

A photo is usually the quickest way to log a plate of real food. It isn't always the most precise one. Pick the method that already knows the answer:

For packaged food, our barcode scanner guide explains how label lookups work and what happens when a product isn't found. For everything else, the beginner's guide to counting calories covers portions by hand, plate and grams.

What should an AI calorie counter app show you?

Whichever calorie counter photo app you try, a few things separate a helpful one from a frustrating one:

• Itemised results. One number for the whole plate is hard to correct. A list of foods with grams is easy.
• A confidence per item. It tells you where to look first: the sauce, not the chicken.
• Editing before saving. Names, grams and whole items should all be changeable without starting over.
• Other ways in. Barcode, search and recents for the moments when a photo isn't the right tool.
• A clear photo policy. You should know where the picture goes, why, and when it's deleted.
• Honest wording. “Estimate”, not “exact”. No app can weigh food through a lens.

If you're comparing options, our guide to choosing a calorie counter app goes through the rest, from barcode coverage to how an app talks to you on a day that didn't go to plan.

What happens while Doy reads your plate?

Doydum has its own camera, so your shot goes straight into a crop step: trim the frame to your plate and, if something is hidden, add a hint. The photo then goes to Google's Gemini model (or to Anthropic's Claude, if Gemini can't answer), only for the estimate you asked for, and the answer comes back in the app's language, English or Turkish.

While you wait, Doy reads the plate in pixels: phone up, flashlight on, shutter, a scan line across the dish, and the macros flying into his phone. Then he reacts to what he saw: a little flex for a protein-heavy plate, a happy savour for something warm and comforting, a zap of energy for anything doughy. The estimate is the same either way; Doy just makes the wait friendlier. The full story is in how Doy reads your plate.

What comes back is a list: each item with grams, calories, protein, carbs, fat and a confidence. You edit what needs editing and save. The photo is kept with the meal you save and deleted with it, and you can scan up to 30 plates a day. There's more about the photo scan on the Doydum homepage and in our privacy policy.

Questions people ask