How Nality's AI estimates work

Nality uses AI in a few places. In each one, the rule is the same: it must be clear what is a database value and what is an estimate.

Photographing a meal

A photo of a meal is sent to an AI model, which names the foods it can see and estimates how much of each there is. The result is labelled AI estimate. It also shows a likely range for the energy, because portion size is genuinely uncertain, and you can edit every item before you save it.

Some things a photo cannot show. Oil and butter used in cooking, sugar in a sauce, and what sits under the top layer of a dish all change the numbers. Curries, pasta sauces and stir-fries are the usual examples. Treat those estimates as a starting point.

Photographing a nutrition label

This is different from a meal photo. Nality reads the values that are printed on the pack. It is not estimating the food, it is reading the text, and you can correct anything it misread.

Describing a meal

When you type or say what you ate, each food you name is looked up in the food databases first, using your own amounts: two tablespoons of peanut butter is converted using your region's tablespoon, and shown as both the spoons and the grams. Only a food that no database has is estimated by AI, and it is labelled the same way. The AI cannot add, remove or swap the foods you named.

What is sent

The privacy policy lists exactly what is sent to the AI provider and what is kept. See the privacy policy.

How much to trust it

• Barcode and database values are the most reliable.
• Label scans are as accurate as the label and the photo.
• Photo estimates are the least precise. Use the range, and edit what you know.

For more on why calorie numbers vary, see how accurate are calorie apps.