The promise is seductive: Take a picture of your meal and artificial intelligence will tell you instantly how many calories you are consuming. No more manual logging. No more guessing portion sizes. No more human error. Cal AI, Lose It! and MyFitnessPal claim that their new photo features will revolutionize calorie counting by letting the smartphone camera do all the heavy lifting.
As someone who has a long and complicated history of calorie counting, and admittedly, an expert in it, I can tell you that calorie counting with a picture is as stupid as it seems.
How AI-powered calorie counters are supposed to work
Calorie tracking apps promise to solve the biggest problem in calorie tracking, which developers claim to be human error. The pitch is compelling: why spend time searching databases or measuring portions when your smartphone can analyze your plate instantly?
Apps that are similar Horse you have Snapcalorie AI uses visual cues such as color, texture and relative size to make educated estimates about what you are eating and how much there is.
They claim AI methods can solve pesky problems of human accuracy when estimating calories–which is fair to say, it’s easy to get wrong. Cal AI is marketed as one of the most sophisticated options available in this space. I decided to test it out myself. The app was free the first three day, and then $29.99/year.
It is easy to set up the app: download it, create an account, enter basic demographic information and set your goals. Here is where I discovered my first red flag. The app informed me cheerfully that “losing 10 lbs is a realistic target”- except that losing 10 pounds actually would push me into the underweight BMI territory. This blanket statement shows a worrying lack of nuance regarding individual health needs.
Cal AI’s photo logging follows these steps:
Ensure that all ingredients are clearly visible and well-lit.
Include an object to scale (like a penny or your hand).
Upload your image and wait for AI analysis.
Review the app’s identifications, and correct any portion estimates. Save the entry in your daily log.
The application provides detailed tips to get better results: Use natural light, avoid shadows and keep the camera parallel with the plate. These guidelines may seem reasonable in theory, yet they reveal the real challenge that these apps face: the complexity of eating in real life.
The reality is wildly disappointed
. I began my testing with a simple thing: a Pink Lady Apple weighing 222 grams. Apples are among the most photographed food items on the planet, and their distinctive color and shape should be easily recognizable.
Cal AI confidently recognized my apple as tikka masla.
Beautiful tikka masala? Meredith Dietz Credit
This time I photographed the apple with its barcode, sitting on a kitchen weight displaying its exact mass. The app recognized it as an Apple this time but estimated that it was 80 calories. The actual count was closer to 120. This is a 33% underestimate, which is not the precision you want when you are trying to track intake accurately.
A more complex meal was the real test: my current meal prepped lunch, which includes fried tofu with onions, cucumbers and tomatoes, feta, chickpeas and a homemade vinaigrette. This is the type of mixed dish which presumably highlights AI’s advantage in comparison to manual logging – no need to search for ingredients or estimate their amounts.
This was a masterclass of algorithmic overconfidence. The app incorrectly identified the golden-brown fried croutons as croutons. I had to correct it manually. The app did a good job of identifying the vegetables and the feta cheese, but completely missed the oil content. The app estimated 450 calories for the entire meal, despite the salad being visibly glistening in dressing.
The estimate was laughably low. One can of chickpeas has approximately My portion contained about 400 caloriesand significant quantities of feta and olive oil-based dressing. This meal should have had a realistic calorie count of 800 to 900.
Its app’s estimation of the portion was even more problematic than the app’s ingredient identification. Cal AI estimated 250 calories when I photographed a smaller portion, less than a quarter the original salad. According to Cal AI’s logic, less that 25% of the original salad contained more than 55% calories. The math doesn’t add up.
Cal AI was way, way off. Credit: Meredith Dietz
This highlights a fundamental limitation of photo-based calorie counting: Cameras capture two-dimensional images of three-dimensional objects. Without consistent reference points or sophisticated depth analysis, estimating volume from photos remains largely guesswork. Even humans struggle with this task, which is why nutrition professionals typically recommend weighing foods for accuracy.
To get a fuller picture of the AI calorie counting landscape, I also tested two other popular apps: SnapCalorie and Calorie Mama.
SnapCalorie: better numbers, same problems
This app has one interesting feature, a “add note” that allows you to provide additional context for ingredients that the camera cannot see. This should theoretically address one of the fundamental shortcomings of photo-based tracking.
SnapCalorie has a useful “add note” feature and more accurate results. Credit: Meredith Dietz
When I tested SnapCalorie with the same Pink Lady apple, it performed much better than Cal AI, estimating 115 calories. But the Greek salad test revealed familiar problems. SnapCalorie’s initial estimate was an absurdly low 257 calories. When I photographed a smaller portion—the same sub-quarter serving that had stumped Cal AI—SnapCalorie estimated 184 calories. The math still didn’t work; this smaller portion should have been roughly 25% of the larger serving, not 70%.
Determined to give the app a fair shot, I used the note feature to manually specify “full container of tofu, feta, chickpeas and olive oil.” With this human intervention, SnapCalorie bumped its estimate to 761 calories—much more reasonable and accurate, though still on the low side.
But this raises the obvious question: If I need to manually input detailed ingredient information to get accurate results, what exactly is the photo accomplishing? I’m essentially doing the work of traditional calorie counting while going through the motions of taking pictures.
What do you think?
Calorie Mama – When AI doesn’t try
Calorie Mama was the most frustrating, and even laughable of the three apps. The interface is rudimentary and the AI performance is so poor, that the app abandons the automated photo analysis premise. Calorie Mama asks you to confirm manually not only the food items, but also their portions after uploading a picture. This defeats the purpose of photo-based logs, as you are doing the same work as manual entry. Calorie mama identified my Greek salad photo as “tofu”- ignoring the vegetables, feta, chickpeas and dressing. The app asked me to manually adjust my portion size, and then seemed to consider logging complete. As if a complex mixture dish was nothing more than plain tofu.
It wasn’t just inaccurate, it was useless. Cal AI and SnapCalorie at least tried to recognize multiple ingredients even if they were off in their calorie estimates. Calorie Mama seemed to have given up on the main challenge, relegating AI to a gimmicky system for storing photos.
AI-powered counting of calories wasted my time.
The promise is efficiency, snap and go. No manual entry needed. My experience revealed a very different reality. I spent a lot of time correcting ingredient names, adjusting portions, and second guessing the app’s estimations. In many cases, I could have done it faster by using a traditional manual logging method with a food weighing scale and database search.
It’s a frustrating dilemma: If you don’t scrutinize the AI’s results, you’ll get wildly incorrect data. If you verify each entry, however, you will lose the time-saving benefits that made the technology so appealing in the first instance. It’s the worst-of-both-worlds – the effort of manual tracking combined the uncertainty of automated guessing.
What’s most concerning is when users lack the background knowledge to recognize inaccurate estimates. My years of calorie counting experience–problematic as that history may be–gave me the knowledge to spot when Cal AI’s numbers were off. What about users who believe in the technology?
A systematic underestimation could be harmful for people who are trying to lose weight as it may lead them to believe that they are eating less than they really do. Overestimation can lead to unnecessary anxiety or restriction around food. Incorrect data undermines tracking in any case.
AI calorie counting apps have a philosophical problem that is not just technical. These tools are based on the idea that precise calorie counting is both beneficial and necessary for health. But Research suggests that obsessive counting of calories may be harmful to many people.
The intuitive eating approach, which focuses more on internal cues of hunger and satiety than external metrics, is a more sustainable, and psychologically healthy, way to approach nutrition. This framework focuses on developing a healthy relationship to food based on the way it makes you feel, rather than hitting numerical targets.
For the majority of people, understanding the general principles of balanced nutritional–eating lots of vegetables, choosing whole grain over refined grains, and including adequate protein- provides better long-term results than meticulous calorie counting.
Bottom line
AI powered calorie counting apps promise that they will solve human error when it comes to dietary tracking. However, they introduce new forms inaccuracy and maintain many of the older problems. These apps may be useful if you’re looking for a rough estimate on how many calories there are in generic foods. For those who want to track their intake with precision, food scales and traditional methods are more reliable.
I would question whether precise calorie-counting serves your health goals. For many people, a more intuitive relationship to food, based on satisfaction and energy levels rather than numerical targets, leads to better mental and physical health. Listening to our bodies may be a better approach than any algorithm.
