By Samipan Bandyopadhyay, developer of Fitto AI
A photo calorie counter does two separate jobs, and it is much better at one than the other.
Job one: what is on the plate
Identifying food from a picture is something modern vision models do well. Rice, grilled chicken, a fried egg, broccoli: clear, separate foods are recognized reliably. Mixed dishes are harder, because a curry or a casserole hides what is in it, but a good model will still name the dish and its likely parts.
Job two: how much of it there is
This is where estimates drift. A photo is flat. It can’t weigh anything, and the same bowl of pasta can look similar at very different weights depending on the bowl, the angle and the distance. Dense foods like rice, nuts and cheese pack a lot of energy into a small area, so a small misjudgement becomes a big calorie difference.
The biggest source of error is usually not the food you can see. It is:
- Cooking oil and butter. A glossy stir-fry or a restaurant steak can carry a lot of fat the camera can’t measure.
- Dressings and sauces, especially on the side or mixed in.
- Hidden layers, like cheese under toppings or rice under a curry.
- Drinks, which often never make it into the photo.
Why an estimate is still worth logging
The point of tracking for most people is awareness: seeing roughly how much you eat, which meals are heavy, and how your week adds up. For that, a consistent estimate you actually log beats a precise number you skip because weighing everything is tedious. The meals people skip logging are often the ones that matter most, like dinners out and takeout.
Five habits that make photo estimates better
- Shoot from slightly above, with the whole plate and something for scale (a fork, a hand) in frame.
- Take the photo before eating, not halfway through.
- Check every portion before logging. In Fitto AI each food gets its own line with a portion you can edit, so one wrong item doesn’t hide inside a total.
- Add what the camera missed: the oil in the pan, the dressing, the drink.
- Use the barcode for packaged food. Printed label data beats any estimate. In Fitto a barcode scan looks the product up in Open Food Facts and costs no credit. If the product isn’t listed, enter the label values yourself.
What an AI estimate is not
It isn’t a lab measurement, and it isn’t medical advice. If you need precise intake for a medical reason, weigh your food and work with a registered dietitian. For building everyday awareness, a quick photo you take at every meal does the job.