The promise is seductive: Snap a photograph of your meal, and synthetic intelligence will immediately inform you precisely what number of energy you are consuming. No extra tedious guide logging, no extra guessing at portion sizes, no extra human error. Apps like Cal AI, Lose It!, and MyFitnessPal’s new photograph options declare to revolutionize calorie monitoring by letting your smartphone’s digital camera do the heavy lifting.
However as somebody with a protracted, difficult historical past with calorie counting—and admittedly, a considerably cursed experience in it—I can inform you that counting energy with a photograph is strictly as silly because it sounds.
How AI-powered calorie counting is meant to work
Calorie counting apps promise to unravel what builders declare is the largest downside with calorie monitoring: human error. The pitch is compelling—why spend time looking databases and measuring parts when your telephone can immediately analyze your plate?
Apps like Cal AI or SnapCalorie AI use visible cues like colour, texture, and relative dimension to make educated guesses about what you are consuming and the way a lot of it there may be.
They declare that AI strategies can remedy the pesky downside of human accuracy in calorie estimation—which, to be honest, is simple to get flawed. Cal AI markets itself as one of many extra subtle choices on this area, so I made a decision to see for myself. The app was free for the primary three days, then $29.99/12 months.
The setup course of is simple: Obtain the app, create an account, enter fundamental demographic info, and set your targets. This is the place I encountered my first crimson flag. The app cheerfully knowledgeable me that “shedding 10 lbs is a sensible goal”—besides that shedding 10 kilos would really push me into underweight BMI territory. This type of blanket assertion reveals a regarding lack of nuance about particular person well being wants.
Cal AI’s photograph logging course of follows these steps:
-
Take a transparent photograph of your meals, ideally in opposition to a plain background.
-
Make certain all substances are seen and well-lit.
-
Embrace a reference object (like a coin or your hand) for scale.
-
Add the picture and watch for AI evaluation.
-
Overview and proper the app’s identifications and portion estimates.
-
Save the entry to your day by day log.
The app gives detailed suggestions for getting higher outcomes: Use pure lighting, keep away from shadows, maintain the digital camera parallel to the plate, and guarantee no substances are hidden. These tips sound affordable in principle, however they trace on the basic problem these apps face—the complexity of real-world consuming.
The truth is wildly disappointing
I began my testing with one thing easy: a Pink Girl apple weighing 222 grams. Certainly this might be a simple win for the AI—apples are among the many most photographed meals on earth, with a particular form and colour that needs to be immediately recognizable.
Cal AI confidently recognized my apple as tikka masala.
Attractive tikka masala, sure?
Credit score: Meredith Dietz
I gave it one other likelihood, this time photographing the apple alongside its barcode and sitting on a kitchen scale displaying its precise weight. The app did acknowledge it as an apple this time, however estimated it at 80 energy when the precise rely ought to have been nearer to 120. That is a 33% underestimation—not precisely the precision you’d need should you’re making an attempt to trace consumption precisely.
The true take a look at got here with a extra advanced meal: my present meal-prepped lunch of fried tofu, onions, cucumbers, tomatoes, feta cheese, and chickpeas, all generously dressed with an oil-based home made French dressing. That is the type of combined dish that presumably showcases AI’s benefit over guide logging—no have to seek for particular person substances or estimate their portions.
The outcomes had been a masterclass in algorithmic overconfidence. The app recognized the golden-brown fried tofu as croutons, which I needed to manually appropriate. It did a fairly good job recognizing the greens and feta, however utterly whiffed on the oil content material. Regardless of the salad being visibly glistening with dressing, the app estimated all the dish at 450 energy.
This estimate was laughably low. A single can of chickpeas accommodates about 400 energy, and my portion included roughly that quantity plus important portions of feta cheese and several other tablespoons of olive-oil-based dressing. A sensible calorie rely for this meal would have been nearer to 800 to 900 energy.
The app’s portion estimation proved much more problematic than its ingredient identification. Once I photographed a smaller serving—lower than 1 / 4 of the unique salad—Cal AI estimated it at 250 energy. In keeping with the app’s personal logic, lower than 25% of the meal by some means contained greater than 55% of its energy. The mathematics merely would not work.
Cal AI was means, means off.
Credit score: Meredith Dietz
This highlights a basic limitation of photo-based calorie counting: Cameras seize two-dimensional photographs of three-dimensional objects. With out constant reference factors or subtle depth evaluation, estimating quantity from pictures stays largely guesswork. Even people wrestle with this process, which is why diet professionals usually advocate weighing meals for accuracy.
To get a fuller image of the AI calorie counting panorama, I additionally examined two different standard apps: SnapCalorie and Calorie Mama.
SnapCalorie: higher numbers, similar issues
SnapCalorie did instantly assuage some skepticism by suggesting a way more affordable day by day calorie goal of 1,900 energy, in comparison with Cal AI’s problematic weight reduction messaging. Nevertheless, this accuracy comes at a steep value—$79.99 per 12 months after only a one-week free trial, making it the most costly choice I examined.
The app does provide one attention-grabbing characteristic: an “add observe” operate that permits you to present extra context about substances the digital camera cannot see. In principle, this addresses one of many basic limitations of photo-based monitoring.
SnapCalorie has a helpful “add observe” characteristic and extra correct outcomes.
Credit score: Meredith Dietz
Once I examined SnapCalorie with the identical Pink Girl apple, it carried out significantly better than Cal AI, estimating 115 energy. However the Greek salad take a look at revealed acquainted issues. SnapCalorie’s preliminary estimate was an absurdly low 257 energy. Once I photographed a smaller portion—the identical sub-quarter serving that had stumped Cal AI—SnapCalorie estimated 184 energy. The mathematics nonetheless did not work; this smaller portion ought to have been roughly 25% of the bigger serving, not 70%.
Decided to provide the app a good shot, I used the observe characteristic to manually specify “full container of tofu, feta, chickpeas and olive oil.” With this human intervention, SnapCalorie bumped its estimate to 761 energy—far more affordable and correct, although nonetheless on the low aspect.
However this raises the plain query: If I have to manually enter detailed ingredient info to get correct outcomes, what precisely is the photograph conducting? I am primarily doing the work of conventional calorie counting whereas going by way of the motions of taking footage.
What do you suppose to date?
Calorie Mama: when AI would not even strive
Calorie Mama offered essentially the most irritating and laughable expertise of the three apps. The interface feels rudimentary, and the AI’s efficiency is so poor that the app primarily abandons the premise of automated photograph evaluation.
After importing a photograph, Calorie Mama requires you to manually affirm not simply the meals objects but additionally their portion sizes. This defeats all the goal of photo-based logging—you are doing all of the work that guide entry would require anyway.
Once I uploaded my Greek salad photograph, Calorie Mama recognized it merely as “tofu”—ignoring the greens, feta cheese, chickpeas, and dressing completely. The app then requested me to manually modify the portion dimension and appeared to contemplate the logging full, as if a fancy combined dish contained nothing however plain tofu.
This wasn’t simply inaccurate; it was ineffective. At the least Cal AI and SnapCalorie tried to acknowledge a number of substances, even when their calorie estimates had been off. Calorie Mama appeared to surrender on the core problem completely, relegating the AI to little greater than a gimmicky photograph storage system.
AI-powered calorie counting wasted my time
The promise of AI-powered calorie counting is effectivity—snap and go, no guide entry required. However my expertise revealed a special actuality. I spent appreciable time correcting ingredient identifications, adjusting portion sizes, and second-guessing the app’s estimates. In lots of instances, I’d have been sooner utilizing conventional guide logging with a meals scale and database search.
This creates a irritating conundrum: When you do not scrutinize the AI’s outcomes, you will get wildly inaccurate information. However should you do confirm each entry, you lose the time-saving profit that justified utilizing the expertise within the first place. It is the worst of each worlds—the trouble of guide monitoring mixed with the uncertainty of automated guessing.
Maybe most regarding is what occurs when customers do not have the background to acknowledge inaccurate estimates. My years of calorie counting expertise—problematic as that historical past could also be—gave me the information to identify when Cal AI’s numbers had been off. However what about customers who belief the expertise?
Systematic underestimation of energy may very well be notably dangerous for individuals making an attempt to drop a few pounds, as it’d cause them to imagine they’re consuming lower than they really are. Conversely, overestimation may trigger pointless restriction or nervousness round meals. Both means, inaccurate information undermines all the goal of monitoring.
The elemental subject with AI calorie counting apps is not simply technical—it is philosophical. These instruments emerge from and reinforce the concept that exact calorie monitoring is each vital and useful for well being. However analysis suggests that obsessive calorie counting might do extra hurt than good for many individuals.
Intuitive consuming, which focuses on inner starvation and satiety cues moderately than exterior metrics, has proven promise as a extra sustainable and psychologically wholesome strategy to diet. This framework emphasizes creating a wholesome relationship with meals primarily based on the way it makes you are feeling moderately than hitting particular numerical targets.
For most individuals, understanding normal ideas of balanced diet—consuming loads of greens, selecting complete grains over refined ones, together with enough protein—gives higher long-term outcomes than meticulous calorie monitoring.
The underside line
AI-powered calorie counting apps promise to unravel human error in dietary monitoring, however they introduce new types of inaccuracy whereas sustaining lots of the previous issues. In case your purpose is solely to get a tough estimate of what number of energy are in generic meals, these apps would possibly present some worth. However for anybody searching for precision of their consumption monitoring, conventional strategies mixed with meals scales stay extra dependable.
Extra importantly, I might query whether or not exact calorie counting serves your well being targets in any respect. For many individuals, creating a extra intuitive relationship with meals—one primarily based on satisfaction, power ranges, and general well-being moderately than numerical targets—results in higher bodily and psychological well being. Possibly the old style strategy of listening to our our bodies works higher than any algorithm.
