← Forky

Photo calorie counting

A photo of the plate, not a search box.

Forky AI photographs a meal, breaks it into components, looks each one up, and writes calories, protein, carbs and fat into your day. This page explains how that works, how close it gets, and the specific cases where it is wrong.

Get Forky on the App Store

Why photo logging exists

Every food diary dies the same way. Not from a lack of motivation — from the fourth day of typing “chicken thigh, roasted, skin on” into a search box and scrolling through nine near-identical entries to pick one. Barcode scanning solved this for packaged food and solved nothing for the food people actually cook.

A photo is two seconds. That is the entire argument. It trades a slice of precision for a large amount of persistence, and for most people persistence is the variable that decides whether tracking does anything at all.

What happens between the photo and the number

The plate is split into components

Rather than asking a model for one number for the whole plate, the image is decomposed: the base layer first (rice, pasta, the protein), then toppings, then sauces and dressings. The three-pass structure exists because single-pass estimates reliably forget the drizzle, and the drizzle is often 200 kcal.

Each component gets a portion estimate

Every identified item is assigned a gram weight from its apparent size on the plate, scaled against the plate and cutlery as reference objects. This is the least certain step in the pipeline and the one you are most likely to want to correct.

Macros come from a lookup, not from the model

Once an item is identified and weighted, its calories and macros come from per-100g nutrition data, not from asking the model to recall how many calories are in chicken. Identification is a vision problem; nutrition is a database problem. Keeping them separate removes a whole class of confident-sounding invented numbers.

You correct it, and the total follows

The result is a list of components you can edit, not a sealed total. Change 150g of rice to 90g and the day's numbers update immediately. Thirty seconds of correction on the two or three foods you eat every week is what moves your log from roughly right to genuinely useful.

How to get more out of it

What Forky does that a pure photo counter doesn't

Photo-to-calories is table stakes now — several apps do it, and Cal AI does it well. Where Forky differs is that the camera also points at the fridge. The fridge scanner builds an inventory of what you own and generates recipes from it with macros attached, so the same app that logs your dinner also had an opinion about what dinner should be. You can also import a recipe from a URL, a photo or a PDF and have its macros computed per portion.

The app is on iPhone only, localised end to end in English, French, German and Spanish, and has a free tier with a daily scan quota so you can test the accuracy claim on your own food before paying anything.

Photo calorie counting questions

How accurate is counting calories from a photo?

Honestly: good enough to steer a diet, not good enough to be a measurement. Whole-plate photo estimates typically drift around ±25%. Decomposing the plate into components and looking each one up per 100g tightens that to roughly ±10–15% on standard plated meals. A 2026 NIH-backed study of four popular calorie apps found all four underestimated real intake by 250–345 kcal per day, which is the honest baseline everyone in this category is working against.

Do I still need a kitchen scale?

For a cutting phase where 100 kcal a day decides the outcome, weigh your food — nothing photographic will match a scale. For everything else the photo is the reason you keep logging at all, and a log you actually maintain at ±12% beats a perfect log you abandon in week three. Many people weigh the two or three foods they eat constantly and photograph the rest.

What does the AI get wrong most often?

Fat you cannot see. Oil absorbed into a stir-fry, butter in a sauce, dressing already tossed through a salad — none of it is visible, and all of it is calorie-dense. Portion depth is the second problem: a photo cannot tell a shallow bowl from a deep one. Both are why gram weights stay editable after the scan.

Can it read a restaurant meal?

Yes, and that is where photo logging earns its keep, because a restaurant plate has no barcode and no label. Expect accuracy to be at the loose end of the range: restaurant cooking uses more fat than home cooking and you cannot see how much. If the dish is obviously oily, nudge the fat component up rather than accepting the estimate.

Does it work on packaged food?

It works, but it is the wrong tool. If the food has a nutrition label, scan the label or the barcode — that is exact data and a photo estimate can only be worse than exact. Photo scanning is for food without a label: cooked meals, restaurant plates, anything you made yourself.

How much does it cost?

The free tier includes a daily AI-scan quota with no card required. Forky Pro removes the cap at $1.99 per week, $5.99 per month or $39.99 per year, with a 7-day free trial on monthly and yearly.