FoodayHow AI counts calories from a photo of your food
An AI calorie counter looks at your photo, recognises each dish, estimates how much is on the plate, and matches that to nutrition data to estimate calories, protein, carbs and fat. The result is a useful estimate you should be able to check and edit, not a lab measurement.
Step 1: finding every dish
The first job is recognition. A vision model looks at the whole photo and decides what is on the table and where. On a busy table that matters: a single photo of a ramen set can hold a bowl of noodles, a plate of fried rice and a small bowl of soup.
When we scanned exactly that photo in Fooday, the AI found three dishes and cut each one out as a sticker: shoyu ramen, fried rice and soup. That separation is what lets you count each dish on its own, or say a dish was shared.
Step 2: estimating the portion
Recognising "ramen" is not enough. A small bowl and a large bowl are very different meals. The model estimates the portion from what it can see: the size of the bowl or plate, how full it is, and the other objects around it.
Portion is the biggest source of error in any calorie estimate, from a person or an AI. That's why a good app shows you the portion it assumed and lets you change it. In Fooday you can switch a dish between a quarter portion and three times the usual size before saving.
Step 3: turning food into numbers
Once the dish and portion are known, the estimate is matched against typical nutrition values for that food. Public databases such as USDA FoodData Central list calories, protein, carbohydrate and fat for thousands of foods, and they're a good place to double-check a number that looks off.
For our ramen photo, Fooday estimated:
| Dish | Estimated kcal |
|---|---|
| Shoyu ramen | 550 |
| Fried rice | 350 |
| Soup | 20 |
| Whole table | 920 |
with 31 g protein, 117 g carbs and 36 g fat for the whole table.
Why the number is an estimate
A photo can't show everything. These are the usual reasons an estimate drifts:
- Hidden fat. Oil in a stir-fry, butter in a sauce or dressing under the leaves adds calories the camera can't see.
- Mixed dishes. Curries, stews and noodle soups vary a lot from one kitchen to the next.
- Portion. A deep bowl looks smaller from above than it really is.
- Leftovers. The app sees what was served, not what you ate.
The fix is simple: treat the number as a starting point. Edit the dish or the portion when you know better, and mark dishes you only had a little of.
How to get a better estimate
- Shoot from above, with the whole table in the frame.
- Use good light and avoid heavy filters.
- Snap before you start eating.
- Check the dish names and portions, and change anything that's wrong.
Is an estimate still useful?
Yes, if you use it the right way. Calorie estimates are most helpful for spotting patterns over days and weeks: which meals are heavy, how often you skip vegetables, how your weekends compare to weekdays. For balanced eating in general, the World Health Organization's healthy diet guidance is a good, plain reference.
Fooday is for everyday wellness, not medical advice. If you have a medical condition or a special diet, talk to a doctor or registered dietitian.