How the AI Meal Scanner
Actually Works

No kitchen scales, no tedious spreadsheets, no endless manual logging. You take one photo of your plate — our multi-modal neural network isolates every single ingredient, approximates its 3D volume, cross-references verified lab databases, and serves up exact macros in under 2 seconds.

< 1.8s

Scan Latency

94.6%

Portion Precision

10,000+

Dishes Trained

Grade A

E-Additive Shield

Live AI Plate Inspector

Select a sample dish below and tap any ingredient to inspect its neural vision breakdown:

Detected Meal Components:

INGREDIENT TELEMETRY

Pan-Seared Salmon Fillet

98.4% match

Calories 360

Protein 36g

Carbs 0g

Fats 22g

✨ Neural Vision Reasoning:

Pan-seared in olive oil. High protein, rich in EPA/DHA Omega-3 fats.

The 6-Step Technology Breakdown

Here is what executes behind the scenes during those 1.8 seconds of scanning:

01

Step 1 • Optics

1. Optical Normalization & Glare Removal

The instant you capture a photo, localized edge-enhancement normalizes ambient shadows, corrects gyro angle, and eliminates flash glare before tokenizing the image.

02

Step 2 • Vision

2. Multi-Modal Deep Neural Vision

Gemini Vision examines surface micro-textures, sear marks, glossiness, grain structure, and browning depth to distinguish fried cutlets from poached chicken.

03

Step 3 • Segmentation

3. Semantic Ingredient Decomposition

Complex meals aren't a mystery box. A double burger is split into individual brioche buns, 80/20 patties, melted cheddar, and secret sauce dressing.

04

Step 4 • 3D Volume

4. 3D Volumetric Depth & Density Math

Using plate geometry (~26cm rim anchor) and perspective depth, the algorithm calculates physical cm³ volume and applies food-specific density coefficients (g/cm³).

05

Step 5 • Verification

5. Verified Nutritional Lab Mapping

Calculated gram weights map directly against USDA FoodData Central and EuroFIR lab tables for complete macronutrients, fiber, sodium, and glycemic load.

06

Step 6 • Guardrails

6. Longevity Shield & Allergen Guardrails

Live checks for personal allergens (gluten, dairy, nuts), dietary inflammatory scoring, AGEs glycation index, and synthetic E-additive hazard detection.

Why Volume ≠ Weight: The Density Simulator

Drag the 3D volume slider and pick a food to see how the formula Weight (g) = Volume (cm³) × Density computes real macros:

3D Spatial Plate Volume: 180 cm³

50 cm³ (Small) 200 cm³ (Standard) 350 cm³ (Large)

CALCULATED MASS & NUTRITION

187g

(180 cm³ × 1.04 g/cm³)

🔥 309 kcal

Protein 58.0g

Carbs 0.0g

Fats 6.7g

💡 Bio Fact: Dense lean muscle tissue with very little air pocket spacing (~1.04 g/cm³).

How to get 99% pinpoint scan accuracy

Simple photography habits that give the neural model razor-sharp precision:

1

The 45-Degree Sweet Spot

A 45-degree angle captures both surface area AND 3D height profile. Flat top-down shots can slightly underestimate steak thickness or layered rice.

2

Include the Plate or Fork as Anchor

Having the plate rim or a fork in the frame gives the AI an instant spatial reference (standard ~26cm dinner plate) to lock in gram calibration.

3

Quick Note for Secret Dressings

If you have heavy dressing buried at the bottom, use the instant voice/text memo ('with 2 tbsp caesar dressing') and the AI recalculates automatically.

4

Barcode & Label Scanner for Packaged Goods

For packaged protein bars, yogurts, and snacks, scanning the barcode or ingredient label yields 100% manufacturer-grade nutritional truth.

Common Questions & Answers

Everything you need to know about our multi-modal neural architecture:

Our 3D volumetric model achieves ~92-96% accuracy on typical whole food plates. For everyday fitness, bodybuilding, and weight management, that 4-8% variance is practically negligible and saves you 15+ minutes of tedious weighing every single day.

The AI recognizes the composite base broth/sauce and detects visible surface chunks (beef, potatoes, carrots, cream). If ingredients are hidden inside, you can mention them in an instant voice note or text memo and the neural engine recalculates the macros immediately.

Our vision model analyzes surface specular highlights (sheen/gloss), char markers (Maillard reaction crust), and color depth. High sheen on chicken breast indicates pan-frying in fat, while a matte pale texture signifies poaching or steaming.

Yes! The AI is trained on extensive global culinary datasets — from local sushi rolls, Neapolitan pizzas, and Kyiv cutlets to fast-food favorites. It accurately predicts restaurant-style higher butter/sodium baselines.

Ready to test it on your next meal?

Snap a photo of whatever is in front of you and get your complete macro breakdown in seconds.

Join Open Beta

How the AI Meal Scanner Actually Works