An AI food recognition app that logs calories from a photo.
Nommie is an AI food recognition app: a vision model identifies the food on your plate, estimates portion size from visual cues, and returns calories and macros. Here's what the technology does well, and where its limits are.
AI food recognition is the core technology behind Nommie. A vision model trained on plated meals looks at a photo, identifies each item, estimates portion volume from visual references, and maps the result to nutrition data. The end product is a calorie and macro log entry — but the underlying recognition step is what makes the rest possible, and it's worth understanding what it can and can't do.
What makes Nommie different
Multi-item recognition
The model identifies multiple foods in the same image — a plate with protein, starch, and vegetables is recognised as three components, not one blob.
Portion estimation
Recognition isn't just 'what is this' — it's also 'how much.' Volume is estimated from visual references like plate size and utensils.
Handles real plates
Trained on actual meals rather than idealized food photography. Steam, bad lighting, messy plates, and mixed dishes are the normal case.
Confidence and fallback
When the model isn't sure, it asks a clarifying question rather than guessing silently. You can also type a description if a photo isn't practical.
AI food recognition is a specific subfield of computer vision. The task is harder than generic object recognition because food categories overlap heavily — grilled chicken and roasted chicken look nearly identical in an image, and the calorie difference between them can still matter. Modern vision models solve this by combining visual features with contextual cues: plate type, cooking method signals like char or sauce, typical meal composition patterns, and user corrections over time.
Portion estimation is the other half of the problem, and it's where most consumer food recognition tools fall down. Knowing an item is 'pasta' is only useful if you can also estimate whether it's a 100-gram side or a 400-gram main. Nommie handles this by using visual references in the image — the plate, utensils, and surrounding objects — as scale anchors. It's not perfect, but it's considerably more reliable than asking users to guess in grams.
There are real limits worth being honest about. Hidden ingredients in mixed dishes (a heavy cream base, a sugar-loaded sauce) aren't always visible in the image. Packaged foods with odd serving sizes benefit from a barcode assist more than pure recognition. And dim or occluded photos reduce confidence. Nommie handles these cases by asking clarifying questions or letting you fall back to a text description — the model is confident when it should be and deferential when it shouldn't.
Frequently asked questions
What is AI food recognition?+
AI food recognition is computer vision applied to meal photos. A model identifies which foods are present, estimates portion size, and — in a calorie-tracking context — maps the result to nutrition data. It replaces manual food-database search with a camera-first workflow.
How does an AI food recognition app estimate calories?+
It combines two steps: identifying each food item in the image, then estimating portion volume from visual references like the plate and utensils. The identified item and portion are mapped to nutrition data to produce calories, protein, carbs, fat, and fiber. Accuracy on real meals is typically within 10-15%.
What are the limits of food recognition AI?+
Hidden ingredients like heavy sauces or cooking fats aren't always visible in an image, so mixed dishes with rich bases can be underestimated. Very dim lighting or heavily occluded plates reduce confidence. Packaged foods with tiny or unusual serving sizes often benefit from barcode or label input in addition to recognition.
How does the app improve over time?+
When you correct an identification or portion estimate, that correction feeds into your personal model. The system learns your specific recipes, your typical portion sizes, and the restaurants you frequent. After a few weeks of use, recognition accuracy on your particular diet is measurably higher than on day one.
Try Nommie for free
Snap a photo, get instant nutrition — no manual logging.