Corvex vs Cal AI

What Cal AI is, and who now owns it

Cal AI is the photo-scanning calorie tracker built by two teenagers that grew to millions of downloads largely through TikTok. Point the camera at a plate, get an estimate. The pitch is that logging should take one second and no typing.

In March 2026 MyFitnessPal acquired it. Cal AI continues as its own app, now with MyFitnessPal’s food database behind the scan. Worth knowing if part of the appeal was backing the independent underdog — that is no longer the shape of the thing.

How good is photo calorie scanning, really?

Photo estimation is genuinely impressive technology and genuinely limited in a way that is easy to talk yourself out of noticing.

It performs best where the food is visible, separable and recognisable — a chicken breast, a bowl of rice, an apple. It performs worst on exactly the things that make meals calorie-dense: oil absorbed into vegetables, butter in a sauce, the dressing already tossed through a salad, anything wrapped, blended, layered or homemade. A camera cannot see the two tablespoons of olive oil the pan absorbed, and that is 240 calories.

Portion size is the second problem. Depth and scale from a single photo are hard, and a 30% portion misjudgement on a 700-calorie plate is worth more than most people’s entire deficit.

None of this makes it useless. It makes it a fast first draft that needs correcting — and the correction step is where the "one second" promise quietly goes.

The retention question

The real difference between these apps is not accuracy on any given meal. It is what happens in month three.

Photo logging is frictionless at the start and can stay shallow: a stream of estimates that never becomes a picture of your eating. Search-and-recents logging is slower on day one and gets much faster as your staples build up — by week three you are tapping "yesterday’s breakfast" rather than photographing it.

Corvex is built for the second pattern, and pairs it with weight trends so the calorie numbers can be checked against reality rather than trusted on faith.

Where Cal AI wins

Straightforwardly, there are cases where the camera is the better tool.

• Unfamiliar food you cannot name, let alone search for.
• Restaurant plates where no entry will ever be accurate anyway and a reasonable estimate beats a guess.
• People who will not type at all — though describing a meal in a sentence, which Corvex supports, is a middle ground worth trying before concluding that.
• The lowest possible friction on day one, which matters if your history is starting and abandoning trackers.
• Android availability — Corvex is iOS-only today.

Where Corvex wins

And the cases where a slower first log pays for itself.

• Repeat meals, which is most meals for most people, logged in a tap from recents.
• Weight tracking with a 7-day trend, so the calorie estimate can be validated against actual weight change.
• No ads on any tier, and a free tier that stays usable.
• Progress stickers designed to be shared rather than screenshotted.
• Estimates you made deliberately, which people trust more over time than estimates handed to them by a model.

A reasonable way to use both

These are not mutually exclusive, and the pragmatic answer for a lot of people is not "pick one".

Use a photo scanner for the meals you cannot otherwise log — the restaurant plate, the friend’s cooking, the thing with no label. Use a habit tracker for the eighty percent of your eating that repeats, and for the weight trend that tells you whether any of the numbers are right. The estimates do not need to be perfect; they need to be consistent enough that the trend means something.

FAQ