Cal AI Alternatives: Calorie Tracker Apps Compared

Cal AI is a photo-based calorie tracker app that estimates calories and macros from a picture of your meal, and it is now owned by MyFitnessPal. Like every photo tracker, it gives an estimate, and research shows those estimates often run low: in a preliminary study by NIH researchers, Cal AI undercounted 102 weighed meals by 345 calories on average, the largest gap of the four apps tested.

Key takeaways

• In a preliminary NIH study, four popular photo apps undercounted meals by 252 to 345 calories, mostly from fat. Cal AI had the largest gap.

• In our tests, Cal AI logged a pork rib soup at 45 to 54 calories across four scans, against 449 to 603 in Dietfit, and a broken rice plate 224 calories lower than Dietfit.

• Dietfit is the only app here with both a fast and a detailed scan, and you can correct it in plain words like "add oil".

What does research say about photo calorie apps?

Photo calorie apps tend to undercount, not overcount, and the gap grows with fat, portion size and mixed dishes. The clearest data comes from a preliminary NIH study that tested four popular apps against 102 meals with every ingredient weighed to 0.1 g.

Source: Charles O, Hengist A et al., NIH National Institute of Diabetes and Digestive and Kidney Diseases, presented at NUTRITION, the American Society for Nutrition's annual meeting. Conference abstract, not yet peer reviewed.

• Fat goes missing. All four apps undercounted fat by about 30 g per meal, roughly 270 calories. Oil and fatty meat add calories without changing much on camera.

• Big portions get underestimated more. In a Nutrients study of 114 meal photos, ChatGPT-4 underestimated medium meals by about a quarter and large meals by about a third.

• Mixed dishes are harder than simple foods. A review of 52 studies in Annals of Medicine found the lowest errors on single or simple foods.

Why do people look for a Cal AI alternative?

Most people who leave Cal AI want more accurate counts and more guidance on what to eat.

• Undercounting on real meals. Cal AI had the largest gap in the NIH study, at 345 calories per meal.

• One scan mode. There is no slower, more detailed mode for soups, curries or stir-fries.

• Numbers, not next steps. Cal AI tells you what a meal contains, but not what to eat next.

Cal AI vs the alternatives: features compared

Every app here now offers some form of AI photo logging, but Dietfit is the only one with two scan modes. The differences are in how much you can correct, whether a coach reads your logs, and what is free.

We scanned the same meals in Cal AI and Dietfit

We logged two dishes that photo apps often get wrong, a Thai pork rib soup and a Vietnamese broken rice plate, using the same photo in both apps. Cal AI came in lower on both, by more than 500 calories on the soup.

Pork rib soup: 54 vs 570 calories

Cal AI named the dish correctly, "pork rib soup hotpot with herbs and chili", but logged it at 54 calories with 2 g of fat. USDA FoodData Central lists braised pork spareribs at 397 calories and 30.3 g of fat per 3.5 oz (100 g). One serving of the ribs in the pot has more than seven times the calories Cal AI logged for the whole dish. Dietfit Deep Scan outlined three parts on the photo, simmered pork 483, spicy pork broth 70 and chopped cilantro 17, for 570 calories and 38 g of fat.

Broken rice plate: 696 vs 920 calories

On a plate of com tam (cơm tấm), the gap was smaller but still 224 calories. Dietfit split the plate into seven parts: grilled pork 286, steamed white rice 260, a second crispy pork item 142, shredded pork 95, soft-boiled egg 78, dipping fish sauce 45 and pickled carrots. Small extras like the fish sauce are easy to skip when you log by hand, and they add up.

Scanning again: consistent vs realistic

We rescanned the soup three more times in each app. Cal AI stayed between 45 and 54 calories every time. Dietfit's four scans ranged from 449 to 603 calories. Dietfit varies more from scan to scan, so it is worth a glance at the parts it found, but every result stayed in a plausible range for a pot of pork ribs. Cal AI was consistent, and consistently low.

What this test does not show. We did not weigh these meals, so Dietfit's numbers are estimates, not lab values. Both apps let you fix a result: Cal AI with its "Fix Issue" button, Dietfit by typing what it missed, such as "add oil" or "less rice". We compared the first answers, because that is the number most people log.

On simple plates, the gap closes. In the Dietfit accuracy benchmark of 30 weighed plates from Google Research's Nutrition5k dataset, Dietfit's estimates were off by 78.9 calories per plate on average and Cal AI's by 80.5. The difference shows up on mixed, fatty dishes like this soup.

The best Cal AI alternatives

Dietfit: best for mixed dishes and knowing what to eat next

Dietfit is an AI calorie tracker app that estimates the calories and macros in your meal from a single photo, then sets a daily calorie goal from your BMR, TDEE and weight goal. Pick Quick Scan for speed or Deep Scan for soups, stews and mixed plates. AI Coach Sara and Meal Score then tell you whether the meal fits your goal and what to eat next, and up to 200 unused calories roll over to the next day to keep a steady calorie deficit on track. It has no barcode scanner yet and no free tier after the 3-day trial. 

If most of your meals don't come with a barcode, download Dietfit and try Deep Scan on your next dinner.

Other options

• MyFitnessPal: best for packaged foods, with more than 20.5 million foods and a barcode scanner. Photo logging and its AI Coach need Premium.

• Lose It!: best free option, with a weight loss plan you can follow without paying. Barcode scanning and device sync need Premium.

• Cronometer: best for vitamins and minerals, with more than 95 micronutrients tracked. Photo logging needs the Gold plan.

• MacroFactor: best for calorie targets that adjust every week to your weight trend. There is no free tier after the trial.

• SnapCalorie: best for portion sizing with the depth sensor on Pro iPhones. Free users get 3 photo logs a day.

FAQ

Is Cal AI free?

Cal AI is free to download, and barcode and text logging are free. Photo scanning, its main feature, needs a paid subscription.

Is Cal AI accurate?

Cal AI is about as accurate as other photo trackers on simple plates, but it tends to undercount real meals. In a preliminary NIH study of 102 weighed meals, Cal AI's estimates were 345 calories too low on average, mostly from missed fat. Every photo tracker gives an estimate, so double-check soups, stews and fried dishes before you log them.

What is the best free alternative to Cal AI?

Lose It! is the best free alternative to Cal AI for most people, because you can log food and follow a weight loss plan without paying. Cronometer is a good free choice if you care about vitamins and minerals.

Can a photo calorie app count mixed dishes?

A photo calorie app can estimate mixed dishes, but the estimate is less reliable than for a single food. Apps that split the dish into ingredients and let you correct each one give you the most control over the final number.

References

• https://www.newswise.com/articles/photo-based-calorie-tracking-apps-may-underestimate-energy-in-meals, Photo-based calorie tracking apps may underestimate energy in meals, American Society for Nutrition news release on the conference abstract by Charles O, Hengist A et al., NIH National Institute of Diabetes and Digestive and Kidney Diseases (preliminary, not peer reviewed)

• https://doi.org/10.3390/nu17040607, O'Hara C, Kent G, Flynn AC, Gibney ER, Timon CM, "An Evaluation of ChatGPT for Nutrient Content Estimation from Meal Photographs", Nutrients, 17(4), 607

• https://pubmed.ncbi.nlm.nih.gov/38060823/, Shonkoff E et al., "AI-based digital image dietary assessment methods compared to humans and ground truth: a systematic review", Annals of Medicine, 55(2)

• https://fdc.nal.usda.gov, FoodData Central, "Pork, fresh, spareribs, separable lean and fat, cooked, braised", FDC ID 167854, U.S. Department of Agriculture, Agricultural Research Service

• Thames Q, Karpur A, Norris W et al., "Nutrition5k: Towards Automatic Nutritional Understanding of Generic Food", CVPR 2021, Google Research