How does QuickCalo calculate calories? Method, data and limits
Last updated: · Prepared by: QuickCalo Team
Short answer
QuickCalo sends your photo or text to an AI model from Google's Gemini family, then checks the result against fixed reference values for portions and drinks. The answer is an estimate, and you can correct the name, portion or calories on the confirmation screen before saving. Your log stays on your phone.
How does QuickCalo analyse a food photo?
- Photo or text: you take a photo with the in-app camera or describe your meal with "Add by typing".
- On your phone: the photo is resized to about 320–512 pixels and re-encoded as JPEG, which removes extra data such as location (EXIF). Grams or millilitres in your text are read separately.
- Encrypted request: the request travels over HTTPS to the QuickCalo server, which runs on Cloudflare Workers.
- AI estimate: the server sends it to an AI model from Google's Gemini family, which estimates the foods, portion type, approximate grams, calories and protein, in your chosen language.
- Reference check: the result is compared with fixed reference values, such as typical portions and drink densities, and implausible values are corrected.
- Cache: the result for the same input is kept for 1 hour so a repeated request gets the same answer. Only the result is cached, never the photo.
- Your confirmation: you review the name, portion and calories, adjust them if needed and save. The log stays on your phone.
How are portions and grams estimated?
The AI compares the space a food takes up with familiar objects such as plates, bowls, glasses and cutlery, and returns a portion type (plate, piece, slice, drink) with an approximate weight or volume. If you type amounts such as "200 g" or "330 ml", they are used in the estimate. On the confirmation screen you can change the portion from half up to two and a half servings, or enter grams directly.
How are the calorie goal and protein reference calculated?
The suggested daily calories start from resting energy use calculated with the Mifflin-St Jeor equation, multiplied by your activity level. To lose weight, up to 500 kcal is subtracted; to gain, 300 kcal is added.
- When losing weight, the suggested deficit is capped at 1% of body weight and at most 1 kg per week.
- The suggestion never goes below 1200 kcal for women or 1500 kcal for men.
- No personal calorie suggestion or goal date is shown for users under 18, during pregnancy, or for goals below a BMI of 18.5.
- The goal date is shown as a range rather than a single day, because everyone progresses differently.
The protein reference is 1.6 g per kg of body weight by default; you can choose 1, 1.4, 1.8, 2 g/kg to match your activity. If your BMI is 30 or higher, the calculation uses the weight that matches a BMI of 25 for your height. You can try the same calculation in the calorie calculator.
What if the result is wrong?
- Wrong food: correct the name and the analysis is refreshed; the first name correction in each analysis is free.
- Wrong portion: change the portion or grams and the calories are recalculated.
- Wrong calories: edit the calories directly.
- Bad photo: retake it; this counts as a new analysis.
What if the food is not recognised?
Even when QuickCalo cannot recognise the food, it usually still returns an estimate and opens the confirmation screen, where you can correct the name or leave without saving. Every analysis that returns a result counts as one analysis. If an analysis fails because of a connection or server error, it does not count.
Known limits of AI calorie estimates
- Hidden oil, butter and sauces raise calories but cannot be seen.
- Mixed dishes and stews: the share of each ingredient cannot be estimated precisely.
- Filled foods such as dumplings, pies and stuffed vegetables hide their filling.
- Poor light or a very low angle can make foods hard to tell apart.
- Depth: food in deep bowls can look smaller than it is.
What happens to my photo?
Your photo is not saved to your gallery or to your meal log. It is resized, stripped of EXIF data and sent only for analysis, and the server does not store it. Details are in the Privacy Policy.
Frequently asked questions
Which AI model does QuickCalo use?
An AI model from Google's Gemini family. Because the model version is updated from time to time, we do not name a specific version.
Is AI calorie counting reliable?
It gives a useful estimate for everyday tracking, but it is not as precise as weighing food and reading labels. The biggest sources of error are hidden oil and sauces, mixed dishes and portion depth, so check each result before saving.
Will the same photo give the same result?
Yes. If exactly the same photo or text is sent again within 1 hour, the server returns the same cached result. A newly taken photo may give a slightly different estimate.
Sources
- Mifflin MD, St Jeor ST et al. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr. 1990;51(2):241–247
- Google Gemini API models — Google's official page on the model family
Download QuickCalo
Free to download for iPhone (iOS 15.0 or later) and Android phone (Android 7.0 or later).