How Accurate Is Photo Calorie Counting?

Photo logging is the reason a lot of people finally stick with tracking. Point, shoot, done, with no scrolling through a database for "chicken breast, grilled, skinless, 4 oz" while dinner goes cold. But a fast number only helps if it's roughly right, so it's fair to ask how close these apps actually get.

The short answer from the research: photo apps are good at recognizing what's on the plate and noticeably worse at knowing how much of it there is, especially the fat. Here's what the studies found, why it happens, and how to get closer numbers without giving up the convenience.

The short answer

• They tend to come in low. In a 2026 NIH study, four popular apps underestimated real meals by 252 to 345 calories on average, about a third of the meal (1).
• Fat is most of the miss. Every app tested missed around 30 grams of fat per meal. At 9 calories a gram, that alone is about 270 calories.
• Simple foods are easier. Across dozens of studies, errors were smallest for single items like an apple and largest for mixed dishes (3).
• People aren't much better. When people estimate their own meals, they tend to come in low by a similar amount (1, 3).

What the 2026 NIH study found

The most direct test so far came from researchers at the NIH's National Institute of Diabetes and Digestive and Kidney Diseases. They prepared 102 meals in a metabolic kitchen, where every ingredient is weighed, photographed them, and compared what four apps estimated with what was actually on the plate (1, 2).

Overall the photo features came in about 33% low. All four missed roughly 30 grams of fat per meal, and a follow-up on more than 200 additional meals suggested the apps do even worse with high-fat, low-carb (keto) meals (2).

Two caveats. The results were presented at a conference and haven't been through peer review yet. And the apps were tested straight from the photo, with nobody adding notes or fixing portions. That's the worst case, not how a careful person logs.

What the wider research says

This isn't a new problem. A 2023 systematic review in Annals of Medicine pulled together studies on estimating food from images (3). Average calorie errors ranged from 0.1% to 38% depending on the study, and the pattern held: errors were lower for single, simple foods and higher for mixed dishes like curries, or plates where foods overlap.

The same review makes a point people tend to forget: human estimates aren't a gold standard either. It cites earlier work where people's estimates ranged from about 30% too low to 1% too high, which puts photo apps in the same range as a person eyeballing the plate. The NIH team said much the same thing: the apps' underestimate looked like what researchers see when people self-report what they ate (1).

Even the app makers are careful about it. Cal AI's own FAQ describes the app as "about 80% accurate" (4).

Why photos come in low

1. Fat is invisible

A tablespoon of olive oil is about 120 calories, and a tablespoon of butter about 100 (5). Cooked into a stir-fry or melted into rice, neither leaves much of a visual trace. Fat has 9 calories per gram against 4 for protein and carbs, so missing a little of it costs a lot. Thirty grams of missed fat, the amount in the NIH study, works out to about 270 calories, which is most of the gap.

2. A photo is flat

A mound of rice could be 150 grams or 300 grams depending on how deep it goes, and a shot from straight above can't tell the difference. Bowls are the worst case.

3. Mixed dishes hide their ingredients

A curry, a casserole or a burrito is a dozen ingredients the camera can't pull apart. That's where the review found estimates drift the most (3).

4. Some food never makes it into the frame

The second helping, the drink, the handful of chips while you cooked. No camera counts what you didn't photograph.

How to get closer numbers from a photo

• Tell the app what it can't see. A short note like "cooked in 1 tbsp butter" or "2 tbsp ranch" fixes the single biggest source of error.
• Shoot the whole plate from an angle. Around 45 degrees shows height as well as area, and a fork in the frame gives a sense of scale.
• Keep sauces and dressings on the side when you can. If the camera can see it, it can count it.
• Scan packaged food instead. A barcode is a lookup, not a guess. We cover where those numbers come from in are barcode calorie scanners accurate?
• Weigh the dense stuff once in a while. Oil, nut butter, cheese, rice and pasta are where eyeballing goes wrong. A week of weighing them recalibrates your eye for months.
• Log home recipes once. For a dinner you make often, log the whole pot and split it by servings, as in counting calories for a family.
• Let your weight check the math. If your weight hasn't moved the way your log predicts after three or four weeks, your entries are probably running low. Nudge your target down a little rather than second-guessing every meal.

How Wellew handles it

We built Wellew around the photo because it's the logging method people actually keep doing. So most of our effort goes into its blind spot:

• Notes go into the estimate. Whatever you type with the photo is sent along with it, so "fried in olive oil" changes the number.
• Re-run with a correction. If a logged meal looks off, add a note like "it was two cups of rice, not one" and Wellew re-estimates from the same photo. Or edit the calories and macros yourself.
• Barcodes are free. Packaged food doesn't need a guess, and scanning doesn't use one of your daily photo scans.

One thing we won't do is quote you an accuracy percentage. Wellew hasn't been independently tested, and a number we made up would be worse than no number. If that changes, we'll publish the results here.

So is photo logging worth it?

Yes, as long as you treat it as a fast, consistent estimate rather than a lab measurement. Research on self-monitoring keeps finding that the people who log consistently are the ones who lose weight, even when the log isn't perfect (6). A photo log that runs a little low every day still shows you which days were heavy and which were light, and your weight trend tells you how much to adjust.

The photo app you'll actually open at dinner beats the precise method you'll give up by Thursday. Just tell it about the butter.