How photo calorie estimation works

A calorie figure that comes out of a food photo is an estimate, not a measurement. Scanning a barcode is a lookup - the number comes from a database record. A photo is not: what you ate is inferred from the image and its nutrition is estimated. That is why Cookrange attaches a confidence score to a photo estimate and shows it to you. "How reliable is this number" is the most-skipped question in the category.

Barcode versus photo, in one sentence

A barcode is a lookup, a photo is an estimate. With a barcode the uncertainty sits in "does a record exist"; where one does, the number is as accurate as the record, andwhere that data comes from is on its own page. With a photo the uncertainty sits inside the number: which dish it is, how much of it there is, and what went into it are each estimated separately. Reading both with the same confidence is the commonest mistake.

Photo analysis produces more than one number

Photograph a meal and, alongside the four core macros (calories, protein, carbs, fat), seven more fields come out: fibre, sugar, sodium, a 0-100 health score, the estimate'sconfidence score, possible allergen warnings and a portion estimate. The health score, the confidence score, the fibre/sugar/sodium values and the allergen warnings appear directly on the scan screen.

What reaches the permanent record - your log - is always just calories and the four macro values. That is deliberate: the rich output on screen belongs to a moment of decision, while what goes into the archive is the minimum that will not distort your daily total. All five logging routes are on the feature page.

Why the confidence score matters this much

A photo estimate is sometimes close and sometimes far off, and the estimate itself does not tell you which. The confidence score fills exactly that gap: a low score means "question this number". A high score is not a guarantee, but a low score is a concrete warning.

If an app shows you only the number and never the uncertainty, it is hiding the interesting part. The same logic applies to a generated recipe's macros; which number in an AI meal plan is arithmetic and which is an estimate isset out on its own page.

Where photo estimates are weakest

• Fat you cannot see. The oil that went into the pan is absent from the photo and present in the calories. Two portions that look identical can differ by hundreds.
• Sauces and mixed dishes. In a stew or on a crowded plate, separating what is there and how much of it from an image is hard.
• Portion depth. A photo is two-dimensional; how full the bowl is, or what is left underneath, does not come out of a single frame shot from above.
• Food stacked on food. What cannot be seen cannot be estimated.

Honest summary: a photo beats not logging a home-cooked mixed dish at all - but for a product with a label, the barcode is always better.

Which route to use when

• Packaged product with a label: barcode. The most reliable route.
• Home-cooked mixed dish: photo. The only realistic option.
• A meal you repeat constantly: quick add. Faster and more consistent.
• A meal from today's plan: pick it from the plan. The number is already computed.

An allergen warning is not a safety guarantee

The possible-allergen warnings from photo analysis are estimates too, and estimates are wrong sometimes. A food allergy is a medical matter: a warning inferred from an image does not replace reading the label or asking whoever cooked the food.Why the allergen filter is a preference filter and what it does not guarantee is on its own page.

What it costs, and where the product stands

Logging by photo is unlimited on the free tier - like barcode and quick add, it does not draw on the daily AI generation quota.Exactly what is in the free tier is written down separately. The app is at v0.9.6 internal alpha and is not in the stores.