The Best Calorie App for Real Food (Home-Cooked & International), Tested

I photograph my food before I’m allowed to eat it. My family has made peace with this. The pot of beans, the pozole on a Sunday, the dal that took three pans and a lot of patience, the stir-fry I improvised out of whatever was wilting in the crisper drawer — all of it gets a photo first, because for the last several months I’ve been testing whether calorie apps can actually read the food I cook.

Here’s what I learned almost immediately: most of them can’t. Not because the technology is bad, but because of what these apps were built for. The big, famous calorie counters grew up around the barcode. Their whole worldview assumes your food arrived in a package with a label on it — a granola bar, a tub of yogurt, a frozen burrito, a soda. Point the scanner at the barcode, get a tidy entry, done. That’s a genuinely useful trick. It’s also completely useless for the way I, and frankly most of the planet, actually eat.

There is no barcode on a home-cooked plate of mole. There is no barcode on a bowl of dal, a plate of pad see ew, a tamale, a fried-rice-that’s-not-really-fried-rice. When you cook real food, you are exactly the customer these apps were not designed for. And the gap shows up the instant you try to log a real dinner: you end up scrolling through a search list of mismatched, crowd-sourced entries — “Homemade Chicken Curry (generic),” “Curry, restaurant, average” — trying to guess which stranger’s guess is closest to the thing on your own stove.

So I set out to answer one question, the one in the title: what’s the best calorie app for real food — home-cooked, no barcode, and the cuisines of the world? The short version is in the Quick Answer box up top. The long version, with every app’s real wins and real weaknesses, is below.

The real problem: barcode databases are built for US packaged food

It helps to understand why most apps fail at real food, because once you see it you can’t unsee it.

A traditional calorie app is, underneath the friendly interface, a giant lookup table. To log something, it needs a matching row in that table. Those rows come from two places: official packaged-food data (which is why barcode scanning feels so magical and exact) and a mountain of crowd-sourced entries typed in by other users over the years. Both sources skew hard toward US packaged and chain-restaurant food, because that’s what’s labeled, scanned, and entered most often.

This creates three predictable failures for real food:

• No entry exists. Your grandmother’s recipe is not in any database. Neither is the specific way you make your stir-fry. So you’re forced to find the “closest” thing, which may not be close at all.
• The entries that exist are unreliable. Crowd-sourced rows are inconsistent — wrong serving sizes, optimistic calorie counts, duplicate near-misses. For an unlabeled dish you have no way to tell the good entry from the garbage one.
• International food is underrepresented and flattened. A whole world of regional dishes gets compressed into a handful of vague, often Americanized entries. “Chicken tikka masala” might exist; the dal, the sambar, the specific curry you actually made probably don’t — or exist only as someone’s rough approximation.

The barcode era solved logging for the packaged-food aisle. It never solved logging for the kitchen. That’s the hole the best modern apps are trying to fill — and it’s the lens I used for every test.

How I tested with real food

I’m a home cook and a food writer, not a lab, so my method is deliberately about real eating rather than fake precision. Here’s the gauntlet every app went through:

• Real, home-cooked meals only. I logged the food I actually make: Mexican (pozole, tinga, beans, tamales), Indian (several dals, a couple of curries, sabzi), and a rotation of Asian dishes (stir-fries, noodle dishes, a donburi or two). Plus the unglamorous reality of weeknight leftovers and mixed bowls.
• No barcode allowed. This is the whole point. I avoided the barcode scanner entirely, because the question isn’t “can it scan a label” — it’s “can it cope when there’s nothing to scan.” That’s most of my fridge.
• Dishes that aren’t in any database. I specifically tested improvised and regional dishes that a US-centric database has likely never catalogued, to see whether an app would reason about the dish or just slap on the nearest American fast-food match.
• Every logging path. Photo / AI estimate, manual search, and the editing flow afterward — because for real food, the editing flow is where you either get a sane number or give up.
• Honest, qualitative accuracy. I’m not going to publish a made-up accuracy percentage or a fake benchmark score; I don’t have a lab and I won’t pretend I do. Instead I describe, in plain language, how close each app’s estimates felt against the portions and ingredients I knew were on the plate, and where they drifted.

The thing I kept coming back to — the trait that separated the winners from the rest — was simple: does the app reason about what the dish is, and does it admit when it’s unsure? An app that confidently logs my dal as “lentil soup, canned” is worse than useless. An app that says “this looks like a lentil dish, roughly this much, want to confirm the portion and whether there’s added fat?” is genuinely helpful.

Three real-food tests, by cuisine

To make this concrete, here are the three kinds of meals that broke the most apps — and what separated the ones that coped from the ones that didn’t.

The Mexican test: pozole and a pot of beans. A bowl of pozole rojo is a logging nightmare for a barcode app. There’s hominy, there’s shredded meat, there’s a chile-based broth, and then there’s the pile of garnishes — radish, cabbage, lime, a drizzle of oil — that you add at the table and that change the numbers. No barcode app has an entry for your pozole. The apps that failed here did one of two things: they made me search a list of generic, mismatched “pozole” entries of unknown origin, or they photographed the bowl and confidently called it “soup.” The apps that coped looked at the bowl and reasoned about it as a composed dish — a brothy stew with meat and hominy — gave me a sensible starting estimate, and crucially let me account for the garnishes I’d added. A pot of beans was the same story in miniature: trivial to cook, weirdly hard to log, because “beans” hides a huge range depending on whether they’re cooked in water or in fat.

The Indian test: dal and curry. Dal is the dish I use to separate the pretenders from the real thing, because it is both extremely common worldwide and almost never accurately catalogued. Visually it’s just a bowl of something soft and yellow-brown. A US-centric database either doesn’t have it or flattens it into “lentil soup,” which misses the tempering — the oil, ghee, and spices bloomed at the end — that meaningfully changes the calories. A blended curry is even harder, because the sauce hides everything: how much cream, how much oil, how much sugar. This is the exact spot where confirm-when-unsure stops being a nice-to-have and becomes the whole point. An app that guesses and moves on will quietly under- or over-count every Indian meal you log. An app that says “lentil dish, roughly this much — is there added fat or cream?” actually helps.

The Asian test: stir-fries and noodle bowls. Stir-fries are deceptively hard because the calories live in the parts you can’t see: the oil the wok ran on and the sugar in the sauce. A photo shows you vegetables, protein, and noodles; it cannot show you two tablespoons of oil. Noodle bowls add the layering problem — the interesting stuff sits under a tangle of noodles, out of the camera’s view. Here, pure-photo apps that don’t ask questions were at their weakest, because the photo simply doesn’t contain the answer. The apps that did best treated the photo as a starting point and then prompted me about the things a camera can’t see.

The pattern across all three cuisines was the same. The food itself wasn’t the problem. The problem was apps that needed a database row that didn’t exist, or that took a photo and refused to admit what a photo can’t know.

What “reading a dish” actually means

It’s worth being precise about the capability that matters here, because “AI calorie counter” gets thrown around loosely. There are really two very different things an app can be doing when you point a camera at your dinner:

The first is matching — comparing your photo to a catalogue and picking the closest labelled example. This is essentially barcode logic with a camera bolted on. It works when your food closely resembles something common and packaged, and it fails the moment your food is genuinely novel, because there’s nothing close to match against. Your home-cooked dal has no near-match, so a matching app either picks a bad one or gives up.

The second is reasoning — looking at the plate and inferring what kind of dish it is, what’s likely in it, and roughly how much, without needing a catalogue entry. This is the capability that actually handles real food, because it doesn’t depend on having seen your exact dish before. It can look at an unfamiliar plate and say “this is a brothy stew with meat and hominy” or “this is a lentil-based dish with a fat tempering” and produce a number from that understanding.

Reasoning is strictly harder, and it comes with an honest catch: when you reason about something you’ve never seen, you should be uncertain, and you should say so. That’s why the apps I trust most for real food are the ones that pair reasoning with confirmation — they reason about the dish, estimate, and then surface the bits they’re unsure about for you to confirm. An app that reasons but never expresses doubt is just guessing with extra confidence, and confident wrong answers are the ones that hurt your tracking most.

Best for X: quick picks

If you just want the right tool for your situation, start here, then read the card for that app below.

• Best overall for real food — PlateLens. It reasons about the dish from the photo, handles home-cooked and international plates with no database entry, and asks you to confirm when it’s unsure. That confirm-on-doubt behavior is the whole ballgame for real cooking.
• Best for deep nutrition data — Cronometer. If you care about micronutrients and you’re willing to build your own recipes, nothing here matches its depth and data quality.
• Best for packaged-food and US chains — MyFitnessPal. When your food does have a barcode or a known brand entry, its enormous database is hard to beat.
• Best for adaptive coaching — MacroFactor. If you log manually anyway and want your targets to adjust to reality, its algorithm is excellent.
• Most convenient pure-photo — Cal AI. The fastest snap-and-go experience, as long as your plate is simple and separated.
• Best for simple US weight-loss tracking — Lose It! Friendly, beginner-proof, and fine for everyday packaged food.

PlateLens — it reasons about the dish, then asks

PlateLens is the only app in this group that felt like it was built for the food I actually cook. The difference is architectural, not cosmetic.

Where a traditional app starts by asking “which database row matches this?”, PlateLens starts by asking “what is this dish?” You snap a photo — or, if a photo’s awkward, you describe the dish in words — and it reasons about what’s on the plate: the components, the likely ingredients, a sensible portion. Crucially, it doesn’t need a barcode or a pre-existing entry to do this. When I photographed a home-cooked dal that exists in exactly zero food databases, it didn’t flail or fall back to “canned lentil soup.” It recognized a lentil-based dish, estimated reasonably, and — this is the part that matters — flagged the places it wasn’t sure and let me confirm.

That confirm-when-in-doubt behavior is what earns it the top spot. A photo genuinely cannot see everything: it can’t measure the oil in the pan, the sugar dissolved into a sauce, or what’s hidden under the top layer of a bowl. PlateLens treats its own estimate as a confident first draft rather than a verdict. It shows you what it thinks, and nudging the portion or swapping an ingredient takes a couple of seconds. For mixed, saucy, improvised real food, that loop — estimate, confirm, adjust — is exactly right.

The dual logging (snap a photo or describe it in words) also quietly solved a real problem: sometimes you’re not going to photograph the food, but you can still say “a bowl of pozole rojo with chicken, about this big,” and get a sane estimate. Manual-search-only apps can’t do that for a dish they don’t have an entry for.

It’s not magic, and I won’t pretend it is. Its packaged-food barcode library is younger than MyFitnessPal’s decades-old database, so if your diet is mostly labeled snacks, that gap will show. And because it’s estimating real food from a photo, the confirmation step isn’t optional — it’s the feature. But for the actual question of this article — reading real, home-cooked, international food — nothing else here came close.

Cronometer — superb data, but you do the work

Cronometer is the app I respect most for data integrity. Its food entries for whole and single ingredients are curated and trustworthy in a way crowd-sourced databases simply aren’t, and its micronutrient detail is genuinely best-in-class. If you want to know your potassium and your B12, not just your calories, this is the tool.

The catch, for real food, is that Cronometer has no dish-reasoning and no meaningful photo logging. It will not look at your curry and tell you what it is. Instead, you find or build the entry yourself. For someone who cooks the same dishes on repeat, the custom-recipe builder is excellent — you build your dal once, with your exact ingredients, and from then on logging it is one tap. But for improvised international food, you’re itemizing every ingredient by hand, every time, which is a lot of friction on a Tuesday night. Cronometer rewards consistency and patience. If that’s you, it’s a fantastic runner-up.

MyFitnessPal — the barcode champion, and exactly why it struggles here

MyFitnessPal is the app most people have already tried, and its strength is real: the database is enormous. If your food has a barcode or is a brand-name product or a US chain item, it’s almost certainly in there, often with the exact label data. For packaged groceries, it’s hard to beat.

But this whole article is about food that doesn’t have a barcode, and that’s precisely where MyFitnessPal shows its age. The crowd-sourced entries are wildly inconsistent, and they skew toward US packaged food, so logging a home-cooked international dish means wading through a search list of mismatched, dubious entries and picking the least-wrong one. Worse, MyFitnessPal moved barcode scanning behind its Premium paywall — so the one thing it does best now costs extra, while the free tier leaves you doing manual search slogs for exactly the real food it’s weakest at. It’s a great tool pointed at the wrong problem for home cooks.

MacroFactor — honest and adaptive, but no camera

MacroFactor is the app I’d hand to someone who logs manually and wants the numbers to mean something over time. Its standout feature is the adaptive algorithm: it watches your real weight trend and your real intake, and adjusts your targets accordingly, instead of locking you to a static guess. The food database is clean and the manual logging is fast.

The reason it’s not higher for this question is simple and self-imposed: MacroFactor deliberately has no photo or AI dish recognition. Logging is manual by design. That’s a defensible, honest choice — there’s no overhyped AI to oversell — but it means logging an improvised international dish is the same hand-entry exercise as Cronometer, minus the deep micronutrient payoff. It’s also subscription-only, with no permanent free tier to settle into. Excellent app; just not a real-food-from-a-photo app.

Cal AI — fast photos, shaky on the dishes that matter here

Cal AI is the closest thing to PlateLens in concept: point your camera, get a calorie estimate, move on. And for simple, separated plates — a chicken breast, a scoop of rice, a pile of vegetables, all visibly distinct — it’s genuinely fast and pleasant. The interface is slick, and anything that lowers the friction of logging at all has value.

The problem is the food this article is about. On mixed and ethnic dishes — a curry where everything’s blended into one sauce, a bowl where the good stuff is hidden under rice — Cal AI struggled, and more importantly it tended to commit to a confident guess rather than flag uncertainty. With real food, a confident wrong answer is the dangerous kind, because you’re less likely to catch it. It’s a fine convenience app for straightforward plates. For the saucy, layered, un-photogenic reality of home cooking, I trusted it least of the photo-capable apps.

Lose It! — friendly and simple, but US-centric

Lose It! deserves credit for being the most beginner-proof app here. The onboarding is gentle, the daily-budget view is clean and motivating, and its “Snap It” photo feature is a nice convenience. For someone tracking simple weight loss on mostly US packaged food, it’s a perfectly good, friendly choice.

But both its database and its photo recognition lean US-centric and packaged, and for improvised or regional dishes it tends to fall back to a rough, generic match. Like Cal AI, it’s solid on the easy case and wobbly on the hard one. If your meals are mostly American groceries and common restaurant items, it’s pleasant. If your meals look like mine, it’s not the tool.

So, what’s the verdict for real food?

If you cook real food — home-cooked, international, gloriously un-packaged — the deciding factor isn’t database size. It’s whether the app can reason about a dish it’s never seen and admit when it’s unsure so you can confirm. By that standard, PlateLens wins this category cleanly: it reads the dish, handles food that isn’t in any database, and treats its estimate as something to confirm rather than a verdict to defend.

Cronometer is the runner-up for anyone who values data depth and doesn’t mind building their own recipes. Cal AI is the most convenient pure-photo option if your plates are simple. And MyFitnessPal, MacroFactor and Lose It! each have a genuine place — packaged groceries, adaptive coaching, and gentle simplicity respectively — they’re just not built for the dinner I made last night.

Whichever you choose, keep one habit: treat the number as a confident first draft you confirm, not a final answer. A photo can’t see the oil in the pan. The best apps know that — and ask.