How it works
How Platey works out what's in a meal
There is no barcode step, no food search and nothing to weigh. Here is what actually happens between photographing your dinner and seeing a number, including the parts most apps do not talk about.
From plate to numbers, in five steps
- 01
Photograph the meal, or describe it
Take a photo of your plate and add a few words if you like — "chicken curry, homemade, low fat". Or skip the photo and type it. Both go to the same place.
- 02
Platey identifies the food and the portion
It works out what the food is and roughly how much of it there is, using the plate, the cutlery, your hands and anything else in the frame for scale. A standard UK dinner plate is about 27cm, which is a surprisingly reliable ruler.
- 03
The estimate is checked against known foods
Around 145 common UK foods are held per 100g. For a plain food — a banana, two eggs, pilau rice — those values are used directly. For a described dish the estimate is kept, but if its energy density is wildly out compared with the known value, it is pulled back towards it.
- 04
The arithmetic is done in code, not by the AI
Meal totals, day totals, what is left against your targets, weekly averages: all of it is plain arithmetic in tested code. The AI identifies food. It never adds up the numbers you see.
- 05
You correct anything that looks wrong
Change a portion, or type exact values off a packet. The correction sticks to that meal, and is recorded so your usual portions can be learned later.
How accurate is it, honestly?
Close enough to be useful, and not as precise as a food label. A photo estimate of a home-cooked curry will not be right to the gram, and no app that lets you skip the scales can promise otherwise.
What it is good at is the shape of a normal week: whether you are reliably short on protein, whether Friday undoes Monday to Thursday, whether the day before a long run looked anything like enough. Those answers do not need gram accuracy, and they are the ones that change what people do.
Three things stop an estimate running away with itself. The AI is asked for structured values rather than prose, so nothing has to be parsed out of a paragraph. Every field is then validated — a food cannot contain more protein than it weighs, portions are held between 1g and 5kg, and the calories are cross-checked against protein, carbs and fat, and recalculated if they disagree by more than a quarter. Finally the reference table acts as a sanity guard on anything that looks implausible for that food.
If you need gram accuracy for a medical reason, weigh your food. Platey is the wrong tool for that and would be a frustrating one.
You do not have to photograph anything
Most logging is typing, and it is meant to be the fast path rather than the fallback. All of these work:
- “handful of chips and a pint”
- “usual breakfast” — once you have saved one
- “two slices of toast with peanut butter, banana, flat white”
- a whole day pasted from memory in one go, which it separates into meals
There is no dropdown of serving sizes and no list of near-identical database entries to choose between. That step is where most food tracking quietly dies.
What it does with the numbers
Your day is shown as one calorie ring and four bars, against targets worked out from your height, weight, age and how active you are. Nothing turns red. Being over a target is information, not a verdict.
Ask what should I eat next and it scores around 45 options plus your own saved and recent foods against the shape of what is left — not just the calories. A day short on protein with little room left gets lean protein; a day where the fat is spent stops being offered nuts. That happens instantly and costs nothing, because it is arithmetic rather than another AI call.
If you run or ride, your targets move around your training — carefully, and without ever handing back the calories you burned.
Easier to try than to read about
Photograph one dinner and see what it makes of it. Setting up takes about a minute.