How to photograph your meal for a better calorie estimate
A clear photo and a few words about what the camera cannot see help an AI calorie counter identify your food and judge portions. Follow these seven simple steps.
The PlateCalories calorie counter works from what it can see. The AI identifies each food and drink in your photo, then estimates portions from a flat image. That is the hard part, because distance and camera angle change how big food looks. These seven steps borrow simple habits from nutrition research, where food photos are used to record what people eat.
1. Use soft, even light
Daylight from a window works well. Avoid strong shadows and bright light behind the plate, and keep your own shadow off the food when you shoot from above. Turn off the flash, which can cause glare, and switch on more lights instead. Research on automatic food recognition lists difficult lighting among its challenges.
Wipe the lens, hold the phone steady and tap the screen to focus. Avoid filters: they change colors, and color helps tell foods apart.
2. Shoot from above, or at 45 degrees when height matters
A top-down photo suits flat meals, because every item is visible. But it hides height and depth: for a bowl of soup, a glass, a burger or a slice of cake, shoot at about 45 degrees.
One food-photo app used in diet research recommends an angle of 45 to 60 degrees, and another system asks for two photos from different angles. PlateCalories analyzes one photo at a time: if you are unsure, take both, send the clearer one and mention what it hides, such as “a deep bowl”.
3. Fit the whole meal in the frame
Include the whole plate with its edges, plus your drink, side sauces, bread and anything else you will eat. It is easy to miss something: in one study, 7 of 78 teenagers trained to use a food-photo app still left a food or drink out of the frame at the first test meal.
Still, let the meal fill most of the picture: PlateCalories shrinks your photo to a maximum of 672 pixels on its longest side before sending it, so small details in a wide shot can get lost.
4. Separate foods when you can
Foods that overlap are harder to identify and measure. When it is easy, spread items out a little and serve sauces on the side. For layered or wrapped foods such as sandwiches, burritos or lasagna, cut one piece open to show the filling, or list the fillings in your description. For a stew or a curry, name the main ingredients.
5. Add a size reference
In research, people often place a reference card of known size and colors next to their food; it helps adjust the images for distance, angle and color. Researchers have also used everyday objects, such as a plate of known size, chopsticks or a thumb.
At home, a fork or spoon next to the plate, your hand beside it or a standard-size plate can give a sense of scale. If you know the size of your plate, bowl or glass, mention it. Do not use a bank card or ID card for scale: it shows personal information.
6. One plate, one photo, before you eat
Take the photo before the first bite. Research methods often add a second photo after the meal to measure leftovers; with PlateCalories, if you do not finish, simply say roughly how much you left. Photograph one plate at a time: your own if dishes are shared, and each course separately.
7. Describe what the photo cannot show
A camera cannot see the oil in the pan, the butter in mashed potatoes, the sugar in your coffee or the filling inside a pastry. Reviews of image-based methods list hidden ingredients and cooking oil among the limits of food photos, and these details count: according to USDA FoodData Central, one tablespoon of olive oil (13.5 g) has about 119 kcal.
Add a voice note (up to 60 seconds) or a written description (up to 500 characters); the AI uses it to correct item names, quantities and hidden ingredients. Mention:
- how the food was cooked (fried, grilled, roasted) and roughly how much oil or butter;
- sauces, dressings, sugar, honey or syrup;
- fillings and anything hidden under other food;
- your drink: regular or sugar-free, whole or skimmed milk, juice or alcohol;
- amounts you know: grams on the package, spoons, slices or pieces.
In a 2025 study of 195 dishes analyzed by another AI model (ChatGPT-5), the average error was about 123 kcal per dish with the photo alone, 92 kcal when details such as the type and amount of fat and sweetener were added, and 53 kcal with a full list of ingredients and amounts. With the same list but no photo, the error was about 67 kcal. Results with PlateCalories may differ, but in that study a photo plus a precise description gave the lowest error.
Vague or helpful? Example descriptions
| Vague | More helpful |
|---|---|
| Salad | Green salad with grilled chicken, half an avocado and 2 tablespoons of olive oil |
| Coffee | Latte, about 350 ml, with whole milk and 2 teaspoons of sugar |
| Pasta | Spaghetti bolognese, about 250 g cooked, with 1 tablespoon of grated parmesan |
| Chicken and rice | Chicken thigh with skin, fried in 1 tablespoon of oil, and about 150 g of white rice cooked with butter |
| Soda | A 330 ml can of regular cola, not sugar-free |
| I did not finish | I left about a third of the rice |
The helpful versions add exactly what a camera cannot see.
Keep your photos private
Only the food needs to be in the picture. Keep faces out of the frame, yours and other people's, especially children's. Move letters, bank cards, ID documents and delivery labels out of the way, avoid screens showing messages, and crop the photo if needed. PlateCalories does not store your photo after producing your result, but it is still wise to share only what is needed.
Check the result
The result lists each food and drink with its estimated portion and calories. If something is missing or wrong, add a correction such as “the glass is white wine, not water” and analyze again (daily usage limits apply). Our guides to portion sizes without a scale and hidden calories can help you describe amounts and extras.
Even a good photo gives an estimate, not a measurement: see our guide to AI photo calorie accuracy. For exact values, weigh your food and check the label. PlateCalories is for adults and is not medical advice. If you are pregnant or breastfeeding, have diabetes or another health condition, or have ever had an eating disorder, talk to a doctor or registered dietitian before changing how you eat.
Sources
- Boushey et al., Proceedings of the Nutrition Society (2017): review of image-assisted and image-based dietary assessment methods
- Höchsmann and Martin, International Journal of Obesity (2020): validity and feasibility of image-assisted dietary assessment
- Chotwanvirat et al., Journal of Medical Internet Research (2024): AI for dietary assessment based on food images
- Daugherty et al., Journal of Medical Internet Research (2012): usability of a mobile phone food record
- Rodríguez-Jiménez et al., Nutrients (2025): energy and macronutrient estimation from food images with ChatGPT-5
- USDA FoodData Central: olive oil, salad or cooking (FDC ID 171413)