FEATURE
Photograph the plate. Verify the lines.
Calorie.one logs meals two ways — a photo, or a plain-English sentence — and both end at the same place: a per-ingredient breakdown you can read, correct, and trust, with every number priced from USDA FoodData Central rather than guessed by a model.
01 · CAPTURE
Snap the plate or type the meal — "leftover pasta with meat sauce and parmesan" is a complete input.
02 · DECOMPOSE
AI splits the meal into ingredients with gram estimates. Recognition is its whole job — not inventing numbers.
03 · GROUND & VERIFY
Each line is priced from a USDA record and labeled with its source. You review, adjust, save.
THE DIFFERENCE
AI that recognizes — a database that counts
Most "AI calorie counter" apps let the model state the calories directly, which produces fluent, untraceable numbers. Calorie.one splits the job at the line where models stop being trustworthy: the AI recognizes ingredients and estimates portions — things vision models are genuinely good at — and the arithmetic comes from USDA FoodData Central, the U.S. Department of Agriculture's lab-analyzed, public-domain food data. Energy is read from USDA nutrient 1008 (kcal); macros come from the same record. Every line carries its source label — usda, mixed, or ai — so an estimate can never masquerade as a fact.
The full pipeline, including how matching works and when a line is demoted to an estimate, is documented on the methodology page.
LINE BY LINE
The preview is the accuracy feature
Before anything is saved, the breakdown is shown for review: each ingredient, its grams, its calories, its source. The model read your bowl as 250 g of rice and it was 180? Edit the line — the per-100 g USDA basis reprices it instantly. This per-line editability is what turns photo logging from a party trick into a method: you correct precisely the part that was wrong, and corrected meals are remembered, so your regulars get more accurate over time, not less.
FAQ
Frequently asked questions
- How does AI photo calorie counting work in Calorie.one?
- You photograph the plate. The AI identifies the ingredients it can see and estimates each in grams. Each ingredient is then matched against USDA FoodData Central and priced from that record — the model never invents the calorie figures. You review the lines, adjust any portion, and save.
- How accurate is calorie counting from a photo?
- Ingredient recognition is very reliable; portion estimation from a photo is honestly approximate — expect roughly ±20–30% per ingredient, tighter for the meal total because line errors partially cancel. That is why every line is shown for review before saving, and why estimates are labeled as estimates.
- Can I log by text instead of a photo?
- Yes — text and photo are equal citizens. "Two fried eggs on rye with butter" produces the same reviewable per-ingredient breakdown as a photo of the plate. Text is often more accurate when you know the amounts; photos win when you don’t want to type.
- Is the photo calorie tracker free?
- Yes. Photo and text AI logging are the core of the free product — not a premium add-on — and there are no ads.
- What happens if the AI gets an ingredient wrong?
- You fix that one line. Every ingredient row is editable before saving — change the food or the grams and the USDA per-100 g basis reprices it instantly. You can also let the AI recalculate a saved meal later.
Point your camera at lunch
Free AI logging by photo or text, grounded in USDA data, with a reviewable breakdown before anything is saved.