What photo estimates get right
Modern AI is usually good at recognising what is on your plate — chicken, rice, salad, pasta — and giving a reasonable starting estimate of calories, protein, carbs and fat.
Where they struggle
- Portion size. A photo is flat, so it's hard to judge depth or weight. This is the biggest source of error.
- Hidden ingredients. Oil, butter, sauces and sugar in dressings don't always show up in a photo.
- Mixed dishes. Stews, curries and casseroles hide what's inside.
- Drinks. A glass of juice and a glass of water can look similar.
Independent tests shared by users online have found individual photo estimates can be noticeably off — often by a fifth or more for a single meal — even when the food itself is identified correctly. Errors tend to even out somewhat across many meals, which is why trends matter more than any single number.
6 tips for better estimates
- Shoot from about 45°, with the whole plate in frame and good light.
- Include something for scale, like a fork or your hand.
- Mention cooking oil, butter or sauces when you edit the entry.
- Adjust the portion if you served more or less than it guessed.
- For packaged foods, use the label numbers instead.
- Look at your weekly trend, not one meal.
How Ash Fuel handles this
Every scan in Ash Fuel is an estimate you can edit. You see the items it found and can change portions or remove items before saving, so your diary reflects what you actually ate. Ash Fuel is a wellness tool — not a medical device — and isn't suitable for medically supervised diets.