The three steps behind every photo calorie tracker
- Recognition. A computer-vision model detects the foods in your photo (chicken, rice, broccoli).
- Portion estimation. The model estimates how much of each food is on the plate — the hardest and least precise step.
- Lookup. Each identified food is matched to a nutrition database (USDA FoodData Central, OpenFoodFacts) and the macros are scaled to the estimated portion.
Cal AI follows this pattern, as does CalorieScan AI and every other photo tracker. The differences are in how each app handles step 2 (portion accuracy) and what it does when the model is uncertain.
Where the apps differ
- Correcting mistakes. CalorieScan AI lets you fix the result in plain English ("no croutons", "double the rice"). When Cal AI is uncertain it may hide the breakdown instead.
- Privacy. Whether your photos are used to train the models — by default or only on opt-in — varies by app.
- Offline. Recognition needs a cloud round-trip; manual entry and previously logged meals usually work offline.
Want the full technical version? We document our entire pipeline on the methodology page.