How Nomvy works out the numbers.

1. Not medical advice

Nomvy gives estimates for general wellness, not medical advice. Equations describe averages across groups of people, and your real needs may differ. Talk to a qualified healthcare professional before changing your diet, especially if you are pregnant, have a medical condition or take medication.

2. Your daily calorie target

Step 1: resting energy (BMR), Mifflin-St Jeor. We estimate basal metabolic rate with the Mifflin-St Jeor equation [1], which a systematic review found to be the most reliable of the common prediction equations for healthy adults [2].

BMR (kcal/day) = 10 × weight (kg) + 6.25 × height (cm) − 5 × age (years) + s, where s = +5 (male), −161 (female), −78 (other or prefer not to say: the midpoint).

Step 2: daily energy (TDEE), activity factor. BMR is multiplied by an activity factor that matches the level you choose [3]: sedentary 1.2, lightly active 1.375, moderately active 1.55, active 1.725, very active 1.9.

Step 3: pace.

• Default pace: 0.5 kg (about 1.1 lb) a week to lose weight; 0.25 kg a week to gain. Maintain uses TDEE as is.

• Maximum pace: weight loss is capped at 1% of your body weight per week [4]. Weight gain is capped at 0.5% of body weight per week, and never more than 0.5 kg a week, to favour lean gain.

• We assume about 7,700 kcal per kg of body-weight change, so the daily adjustment = weekly pace (kg) × 7,700 ÷ 7. Target = TDEE − adjustment (lose), TDEE + adjustment (gain) or TDEE (maintain).

Step 4: calorie floor and rounding.

• The target never goes below 1,200 kcal/day for women, or 1,500 kcal/day for men and for anyone who chose "other" or did not specify, in line with long-standing guidance that lower intakes need medical supervision [5]. If the floor applies, the app tells you and slows the projected pace accordingly.

• The result is rounded to the nearest 10 kcal.

• Custom targets you set later (a Pro feature) can't go below the same floors.

Step 5: projected goal date. If you set a target weight, the app projects a goal date from the effective pace after the floor: weeks = weight to change ÷ effective weekly pace. It's a projection, not a promise. Real weight change is not linear, and the 7,700 kcal/kg rule is a simplification [6].

3. Safety rules

• Age: you must be 16 or older. Sign-up is blocked for anyone younger.

• Low BMI: if your BMI is under 18.5, or your target weight would put your BMI under 18.5 [7], the weight-loss goal is unavailable and the app shows support resources (NEDA, Beat, findahelpline.com).

• Pace caps and calorie floors as described above, applied to calculated and custom targets.

• Gentle mode lets you hide calorie numbers and see rings and colours instead.

4. Protein, fat and carbs

Protein is set in grams per kg of your goal weight (your target weight if you're losing or gaining, otherwise your current weight), within the ranges of the International Society of Sports Nutrition position stand [8]: lose weight 1.2 g/kg, maintain 1.4 g/kg, gain weight or build muscle 1.6 g/kg, and at least 2.0 g/kg with the "High protein" diet preference.

• Default: fat provides 30% of calories (paleo: 35%), and carbohydrate takes the remaining calories after protein and fat.

• Low-carb preference: carbohydrate is 25% of calories; fat takes the remainder.

• Keto preference: carbohydrate is 5% of calories, kept between 20 and 30 g; fat takes the remainder.

• Fat never falls below 25% of calories, within the Acceptable Macronutrient Distribution Range of 20–35% [9]. If protein would crowd fat below that minimum, carbohydrate is reduced first (to no less than 0 g) and then protein is trimmed.

• Energy is converted with the Atwater factors: 4 kcal/g for protein and carbohydrate, 9 kcal/g for fat [10]. Values are rounded to whole grams.

5. Fiber, water and steps

• Fiber: 14 g per 1,000 kcal [9], with a minimum of 25 g a day (keto: 25 g).

• Water: 35 ml per kg of body weight, rounded to 50 ml and kept between 1.5 and 4 litres a day. This is a convenience target. Food and other drinks also count toward total water intake [11].

• Steps: 6,000 (sedentary), 7,500 (lightly active) or 8,000 (moderately active and above).

6. Worked example

A 30-year-old woman, 165 cm, 70 kg, moderately active, who wants to reach 62 kg at the default pace:

• BMR = 10×70 + 6.25×165 − 5×30 − 161 = 1,420 kcal

• TDEE = 1,420 × 1.55 = 2,201 kcal

• Pace = 0.5 kg/week (cap: 1% × 70 kg = 0.7 kg/week), so 0.5 × 7,700 ÷ 7 = 550 kcal/day

• Target = 2,201 − 550 = 1,651, rounded to 1,650 kcal (above the 1,200 kcal floor)

• Protein = 62 kg × 1.2 = 74 g · Fat = 30% = 55 g · Carbs = remainder = 214 g

• Fiber 25 g · Water 2,450 ml · Steps 8,000 · 8 kg at 0.5 kg/week ≈ 16 weeks

7. How AI food estimates work

• Input. Photos are resized (1,024 px on the long edge) and stripped of location and other metadata on your phone. Voice is transcribed on your device. Text and photos are sent to a vision-language model (Claude, by Anthropic).

• Recognition. The model is instructed to identify every dish and component, estimate each item's weight from visual cues (plate size, utensils, piece counts) in natural household units, and give calories, protein, carbs, fat and fiber, a confidence for each item, and an overall likely range. Harder meals (low confidence, or four or more items) are re-checked with a larger model.

• Hidden calories. The model must list invisible contributors (cooking oil, ghee, butter, dressings, sauces, sugar or milk in drinks, fried coatings) as separate lines, using cuisine-aware defaults. When it's unsure, it asks at most one question. For "How was this cooked?" the answer is applied with a fixed table, scaled to the portion: no oil 0, a little oil 1 tsp, restaurant-style 1½ tbsp, ghee/butter-rich 2½ tbsp.

• Database grounding. Each item name is matched against your custom foods, our regional dish pack and USDA FoodData Central. When a match is close enough, the database's nutrition per gram replaces the AI's numbers, while the AI's portion estimate is kept. Matches whose energy density differs implausibly from the AI's estimate are rejected, and we avoid counting cooking fat twice when a database dish already includes it.

• Consistency checks. If an item's calories disagree with its macros (4/4/9) by more than 30%, the macros win. All values are clamped to plausible limits, and the range is rescaled so it always brackets the total.

• Smart Memory. When you correct portions, Nomvy learns a personal ratio for that food (an exponentially weighted average). It applies only after at least two corrections, is limited to 0.5–2×, is never applied to hidden-calorie lines, is always labelled ("Adjusted for you"), and can be switched off or reset.

• You review. Results are labelled "AI estimate". You can edit grams, answer the question, or fix the meal with a sentence. Corrections are applied as specific operations on named items, so other items can't change by accident. Only the point estimate counts toward your daily total; the range is for information.

Quality control. Before we change a model or prompt, we re-test it on a set of meals weighed in a kitchen, including South Asian and Middle Eastern home-cooked dishes, and on a set of correction sentences. We check that estimates are not systematically low. We don't publish an accuracy percentage, because accuracy varies widely from meal to meal and a single number would be misleading.

Why this matters. A 2026 controlled-feeding study presented at NUTRITION 2026 found that four popular photo-based apps underestimated calories by about 33% (roughly 250–345 kcal per meal), mainly from fat [12]. The Hidden Calories Check, ranges and one-sentence corrections are designed around that failure mode.

8. Limits of estimates

• A photo can't show what's inside a dish (fillings, stuffing, sauces underneath), how much oil was absorbed, or what's in a drink.

• Portion size from a single 2D photo is uncertain, especially for mixed dishes, bowls and food piled high.

• Recipes vary between households and restaurants. Database values are typical figures, not your exact meal.

• Packaged foods are more reliable when scanned by barcode, but Open Food Facts data is crowd-sourced and can contain errors.

• Prediction equations such as Mifflin-St Jeor estimate resting energy within about 10% for most healthy adults, but can be further off for individuals, including people with very high or low muscle mass [2].

• Nomvy is not suitable for allergen checking or for medical decisions such as insulin dosing.

9. Data sources

• USDA FoodData Central: Foundation Foods, SR Legacy and FNDDS for generic foods, and Branded Foods as a barcode fallback [13]. Public domain.

• Open Food Facts: barcode and packaged-product data, used under the Open Database License (ODbL); individual contents under the Database Contents License [14].

• Nomvy regional dish pack: a curated set of South Asian, Middle Eastern and other home-cooked dishes and basic ingredients, with per-100 g nutrition and household serving sizes (katori, plate, piece, roti, tablespoon) and local names. Values are typical home-preparation figures compiled from USDA FNDDS, the Indian Food Composition Tables 2017 [15] and published recipe analyses, and each entry is checked for Atwater consistency.

• Your own data: custom foods, recipes and saved dishes you create.

10. References

[1] Mifflin MD, St Jeor ST, Hill LA, Scott BJ, Daugherty SA, Koh YO. A new predictive equation for resting energy expenditure in healthy individuals. Am J Clin Nutr. 1990;51(2):241–247. doi:10.1093/ajcn/51.2.241

[2] Frankenfield D, Roth-Yousey L, Compher C. Comparison of predictive equations for resting metabolic rate in healthy nonobese and obese adults: a systematic review. J Am Diet Assoc. 2005;105(5):775–789. doi:10.1016/j.jada.2005.02.005

[3] FAO/WHO/UNU. Human energy requirements: report of a joint FAO/WHO/UNU expert consultation. Rome, 2004 (physical activity level multipliers).

[4] Helms ER, Aragon AA, Fitschen PJ. Evidence-based recommendations for natural bodybuilding contest preparation: nutrition and supplementation. J Int Soc Sports Nutr. 2014;11:20 (0.5–1% body weight per week). doi:10.1186/1550-2783-11-20

[5] National Heart, Lung, and Blood Institute. Clinical Guidelines on the Identification, Evaluation, and Treatment of Overweight and Obesity in Adults. NIH Publication 98-4083, 1998 (1,000–1,200 kcal/day for women, 1,200–1,600 kcal/day for men).

[6] Hall KD, Sacks G, Chandramohan D, et al. Quantification of the effect of energy imbalance on bodyweight. Lancet. 2011;378(9793):826–837. doi:10.1016/S0140-6736(11)60812-X

[7] World Health Organization. A healthy lifestyle: WHO recommendations (BMI classification).

[8] Jäger R, Kerksick CM, Campbell BI, et al. International Society of Sports Nutrition position stand: protein and exercise. J Int Soc Sports Nutr. 2017;14:20. doi:10.1186/s12970-017-0177-8

[9] Institute of Medicine. Dietary Reference Intakes for Energy, Carbohydrate, Fiber, Fat, Fatty Acids, Cholesterol, Protein, and Amino Acids. National Academies Press, 2005. doi:10.17226/10490

[10] Merrill AL, Watt BK. Energy Value of Foods: Basis and Derivation. USDA Agriculture Handbook No. 74, 1973 (Atwater factors).

[11] EFSA Panel on Dietetic Products, Nutrition, and Allergies. Scientific Opinion on Dietary Reference Values for water. EFSA Journal. 2010;8(3):1459. doi:10.2903/j.efsa.2010.1459

[12] Healio. AI photo-based calorie-tracking tools underestimate calories by 33% (NUTRITION 2026). 4 August 2026.

[13] U.S. Department of Agriculture, Agricultural Research Service. FoodData Central. fdc.nal.usda.gov

[14] Open Food Facts. world.openfoodfacts.org; Open Database License: opendatacommons.org/licenses/odbl/1-0/

[15] Longvah T, Ananthan R, Bhaskarachary K, Venkaiah K. Indian Food Composition Tables 2017. National Institute of Nutrition (ICMR), Hyderabad.