How to Track International Food with an AI App

To track international food with an AI app, photograph the dish before you eat, add its name in a short note ("chicken tikka masala, restaurant"), check the portion, and add any oil, ghee or cream the camera can't see. For reference, USDA lists a cup of dal at 348 calories and a cup of chicken curry at 257, before the rice or bread.

When the app can't place a dish, log it by its parts: starch, protein, sauce and fat. That works for any cuisine.

Key takeaways

• Name the dish. A one-line note with the dish name and how it was cooked gives the AI far more to work with than the photo alone.
• Portion is the big error. In the Nutrition5k study, photo-only calorie estimates were off by about 26%, versus about 9.5% when the weight was known.
• Hidden fat comes next. One tablespoon of oil or ghee adds about 120 calories, and restaurant curries, stir-fries and fried rice can use several.
• Log by parts when stuck. Rice + protein + sauce + oil beats a wrong database match.
• Save what you repeat. Your regular pho or dal only needs careful logging once.

Why international dishes trip up calorie apps

Database apps match the words you type to an entry someone created. That works well for packaged food and big US chains. It works less well when:

• The dish has several names or spellings (doro wat, doro wot, doro wet).
• The same name covers very different recipes. A cream-and-butter restaurant korma and a thin home-style chicken curry can both be logged as "chicken curry".
• The recipe is the restaurant's own, with no published nutrition.
• Most of the calories sit in fat you can't see: ghee in dal, coconut milk in a Thai curry, oil in fried rice.

A photo-based tracker gets around the naming problem because it looks at the food, not the word. It doesn't solve the hidden-fat problem on its own. That's why the note and the portion check matter.

How to track any cuisine with an AI app

1. Photograph the dish before you eat

Shoot from directly above with the whole plate or bowl in frame, in decent light. For shared plates, photograph the full spread before anyone serves themselves. Our guide on how Nutrify identifies dishes from any cuisine explains what the AI looks for.

2. Name the dish and how it was made

A few words do a lot:

• "Nigerian jollof rice with fried chicken"
• "Tonkotsu ramen, restaurant, half the broth left"
• "Injera with misir wat and gomen, shared between two"

3. Check the portion

This is the step that moves the number most. Compare the estimate with what's actually on your plate. If you ate half a shared platter, log half.

4. Add the fat the camera can't see

Oil, ghee, butter and cream disappear into sauces. If the dish looks glossy or the curry is rich, add a tablespoon or two. Each tablespoon is roughly:

5. Save it

Once the entry looks right, save it as a meal. Next time you order the same dish from the same place, logging takes a tap.

Calories in common international dishes

These are USDA survey (FNDDS) values for typical home or restaurant versions. Real dishes vary, so treat them as a starting point to compare against the AI's estimate, not as the answer.

Two things stand out. Soups like pho and miso are low per 100 g, but restaurant bowls hold several cups, so the bowl size decides the total. Rice, noodle and wrap dishes are dense, so a "small" difference in portion is a big difference in calories.

Cuisine-by-cuisine tips

Indian

Tandoori dishes, plain dal and rice are the easiest to log. Cream and ghee curries such as butter chicken and korma are the hardest, so estimate those on the high side and log rice and naan as separate items. More numbers in our Indian food calories guide.

Japanese

Nigiri, sashimi and rolls come in standard sizes, so counting pieces works well. Ramen is the tricky one: the broth carries a lot of the fat, so note the style (shoyu and shio are lighter than tonkotsu) and how much broth you drank. See the Japanese food nutrition guide.

Mexican

Street tacos are small and easy to count. Burritos and combination plates hide rice, beans, cheese and sour cream inside, so describe the fillings in your note. Our Mexican food nutrition guide covers common orders.

Korean

Bibimbap photographs well before you mix it. For Korean barbecue, log each meat as you order it and add the banchan you actually eat.

Chinese, Thai and Vietnamese

Dim sum is easiest one basket at a time, photographed as it arrives. Stir-fries, fried rice and noodle dishes such as pad Thai carry most of their calories in oil and sauce, so add a tablespoon of oil when the dish looks glossy.

Ethiopian and West African

Photograph the shared platter before anyone starts, then log your share as a fraction (half for two people, a third for three). Log the injera or fufu as its own item. Our Ethiopian food nutrition guide lists common wats.

When the AI gets it wrong

Log it by parts

Break the dish into what it's made of:

• Jollof rice: cooked rice + tomato sauce + oil + the chicken on the side.
• Pho: broth + rice noodles + beef slices + herbs.
• Chicken with mole: chicken + mole sauce.

Each part has a reliable entry even when the dish doesn't.

Use the closest USDA match

USDA's survey database includes many global dishes (dal, fufu, bibimbap, pho, pad Thai, mole). If your app lets you search, a USDA-based entry for the dish is a reasonable fallback, adjusted for portion.

Weigh it once

For a dish you eat every week, weigh one serving once. Knowing that your usual takeout curry container holds about 2 cups makes every later estimate better.

How accurate is "good enough"?

You don't need perfect numbers. You need numbers that are consistent enough to show the trend. A 15% miss on a 350-calorie bowl of dal is about 50 calories, small next to the day-to-day swings in your appetite and activity. The misses that matter are the big ones: a forgotten tablespoon of ghee, or logging one cup when you ate three. Fix those two and international food is no harder to track than anything else. For more on where photo estimates go wrong, read how accurate is AI calorie counting.

Tracking international food in Nutrify

We make Nutrify, so here's how this works in our app. After you take or pick a photo, you can add a typed or spoken note (the dish name, how it was cooked, how much you ate) before the analysis runs. On the result you can change the name, portion and each ingredient, then save the meal to reuse. The app is available in 15 languages, and you can also log by barcode, Siri or a typed description. Nutrify has a 7-day free trial, then a paid subscription.

Sources

• USDA FoodData Central, Food and Nutrient Database for Dietary Studies (FNDDS): chicken curry, FDC 2706437; dal, FDC 2707427; bibimbap, FDC 2708950; pho with meat, FDC 2707124; pad Thai with chicken, FDC 2708806; ramen bowl, FDC 2709153; California roll, FDC 2708961; steamed dumpling, FDC 2708708; meat burrito with beans and rice, FDC 2708553; fufu, FDC 2709568; chicken with mole sauce, FDC 2706439; fried rice with chicken, FDC 2708953; miso soup, FDC 2707455; meat taco, FDC 2708530.
• USDA FoodData Central, SR Legacy: olive oil, FDC 171413; sesame oil, FDC 171016; ghee, FDC 171314; butter, salted, FDC 173410; canned coconut milk, FDC 170173.
• Thames Q, et al. Nutrition5k: towards automatic nutritional understanding of generic food. CVPR 2021.
• Apple App Store, Nutrify - AI Calorie Tracker: logging methods, languages and subscription terms. Checked October 26, 2026.