Photo logging is the reason people finally stick with tracking. It is also the feature most likely to be quietly wrong. Here is an honest look.
Where estimates go wrong
- Hidden calories. Oil, butter, dressings and sauces are invisible or nearly so.
- Portion depth. A bowl of rice looks the same from above whether it is 150 g or 300 g.
- Look-alikes. Full-fat vs zero-fat yoghurt, diet vs regular cola.
A pure “AI guess” approach compounds all three. The model estimates calories directly from pixels with nothing to anchor it.
What database grounding changes
Vitalume recognises foods from the photo, then looks each one up in a verified nutrition database and estimates the portion separately, using depth cues and plate references. The calories come from real lab data, not from the model’s impression. When it is unsure, it says so and asks.
Five habits that get you within 5%
- Shoot from a 45° angle, not straight down; depth is what portion estimates need.
- Mention the sauce. “Chicken rice bowl, with the teriyaki” is a two-second caption that fixes the biggest error.
- Use barcodes for packaged food. Exact beats estimated.
- Correct once, benefit forever. Every correction updates your personal portion priors for that food.
- Weigh your staples once. Know what your usual oats or rice serving weighs and the estimates snap to it.
The honest summary
Photo logging with grounding and light captions lands within 5–10% for most meals, comparable to careful manual logging, at a fraction of the effort. And because targets in Vitalume adjust to your real weight trend, a consistent small error washes out within two weeks.