Food Recognition API

Photo to nutrition in about a second

Food logging lives or dies on friction. This model is tuned for food and runs on a low-latency inference stack, so a full analysis comes back in about a second — against the 4–10 seconds a general-purpose multimodal model needs for the same photo. The result lands before the user loses interest, and the camera flow never stalls on a spinner.

That speed also makes batch work practical: catalogs of dish photos, moderation queues, and bulk backfills process without per-image latency dominating the run. See the benchmarks for the head-to-head.

From a photo to structured nutrition

Most "food recognition" returns only a label — you still have to map that label to nutrition yourself. TastyAPI does both in one request: it recognizes the dish and returns the nutrition for the portion it sees, so a single call is enough to log a meal.

Quickstart

Upload a file as multipart form data, or pass a public image URL. Authenticate with a bearer token.

curl -X POST https://tastyapi.com/analyze-image \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "image=@lunch.jpg"

{
  "analysis": {
    "foodName": "Margherita Pizza",
    "servingSize": { "amount": 107, "unit": "g" },
    "nutrients": {
      "calories": { "amount": 285, "unit": "kcal" },
      "protein": { "amount": 12, "unit": "g" },
      "carbohydrates": { "amount": 36, "unit": "g" },
      "fat": { "amount": 10, "unit": "g" }
    },
    "allergens": { "contains": ["gluten", "dairy"], "mayContain": [] },
    "dietaryInfo": { "isVegetarian": true, "isVegan": false, "isGlutenFree": false },
    "imageAnalysis": { "quality": 0.94, "multipleFoods": false, "needsClarification": false }
  },
  "language": "en"
}

What it handles

• Prepared meals and plated dishes — the everyday photos people actually take of their food.
• Packaged products and raw ingredients — from a snack wrapper to a single piece of fruit.
• Portion estimation from context — nutrition scales to the amount visible, not a fixed 100 g.
• Allergen and dietary flags — gluten, dairy, and vegan/vegetarian/gluten-free booleans in the same response.

Where it fits

• Photo-based food logging. Replace manual search-and-select with a single snapshot — the highest-friction step in any diet app.
• Fitness and coaching apps. Auto-fill macros for a meal from the photo a user already shares.
• Restaurant and delivery. Generate nutrition estimates for dish photos across a catalog.
• Health assistants. Add a vision "what am I eating?" capability to a chatbot or agent.

Accuracy and image tips

The model is trained on tens of millions of food images and returns calibrated estimates, typically within 10–15% of laboratory analysis. Recognition is strongest on clear, well-lit photos where the whole dish is visible. Heavy occlusion, extreme angles, or very mixed plates lower confidence — for those, a follow-up text call to the Nutrition API with a short description can refine the result. Estimates are for tracking and information, not medical use.