Photos often miss oil, sugar, sauces, and fillings not because AI failed to recognize them, but because these ingredients have blended into the food, are hidden below the surface, or never appear in the frame. Glossy braised pork suggests sauce, but not how many spoonfuls of oil and sugar were used. An intact bao does not reveal its filling.
nosh can recognize food from a photo and estimate calories, but unseen ingredients still require corrections from information you know. The important distinction is not how finely you can guess. It is which clues exist in the image and which answers can only come from a recipe, package, or cook.
Oil can remain in the pan—or quietly enter the food
A plate of stir-fried greens may look light without being low in oil. A photo cannot reveal how much oil fried eggplant absorbed from its shine alone. Cooking oil may coat the surface, soak into ingredients, remain at the bottom of the plate, or blend into the broth.
An image can show signs of oil, but converting those signs into exact grams is difficult. The same dish may be prepared very differently at home, in a cafeteria, or at a restaurant. Even with the correct dish name, oil may remain one of the meal’s largest uncertainties.
If you know the dish was cooked with little oil or left most of the oily liquid behind, preserve that fact. If you do not know, do not invent “one spoonful” from its color.
Sugar and sauces often have no visible boundary
Once sugar dissolves into braising liquid, sweet-and-sour sauce, or a drink, the photo does not show a small pile of white crystals. Sesame paste, salad dressing, and hot-pot dipping sauce may cover the food. AI can see that a sauce is present without knowing its concentration, recipe, or how much you consumed.
One person may use a full packet of sesame sauce on cold noodles while another uses half. Two coffees with a similar color may be an unsweetened latte and a syrup-sweetened drink. Similar appearances do not prove identical recipes.
For a separate sauce packet, the labeled amount and the portion you used provide more information than the image. “No sugar,” “half sugar,” or “sauce on the side” from the order can also help. “It does not look sweet” is not dependable data.
Fillings are different: the camera never saw them
Bao, dumplings, rice balls, sandwiches, and filled breads hide important information below an outer layer. A white bao may contain pork, red bean paste, or vegetables while looking almost identical from outside. A cut mooncake reveals the filling; an intact one requires flavor or package information.
Another angle cannot always solve this. Include a cut surface when convenient. If you do not want to open food for the sake of logging, add the name, package details, or menu information instead. Eating should not have to obey the camera.
Use information in a sensible order
For hidden ingredients, look for evidence in this order:
- Ingredients, net weight, and nutrition facts on the package;
- Ingredients and amounts you actually used when cooking;
- Restaurant menus, flavor choices, and separate sauce packets;
- When only a photo exists, keep the result as a rough estimate.
The principle is simple: do not ask a photo to guess what is already written or measured, and do not hide missing information behind a specific number.
In nosh, add what you know—not what you wish you knew
If a pork-and-cabbage bao is identified as plain mantou, edit the name. If you used half a packet of sauce, adjust the portion. When the photo is missing or cannot show the preparation, log it with text or voice.
nosh lets you edit names, calories, portions, and log times. Prioritize details that change the estimate meaningfully: fried, sweetened, meat-filled, or half the sauce used. If you have no basis for whether a restaurant used 18 or 25 grams of oil, leave that uncertainty intact.
The photo is not lying; it simply did not see everything. A useful food log knows when to look at the image, when to read the label, and when to accept that a meal can only be recorded approximately.