Food analysis

Every call on this page needs the food_analysis:write scope. An analysis can take tens of seconds, so show progress.

Analyze a description

const scan = await january.foodAnalysis.analyzeDescription({
  query: 'two eggs, avocado toast, and black coffee',
});

analyzeDescription takes only query.

Analyze a photo

analyzePhoto takes a remote image URL or a base64 data URI. The SDK has no camera UI; this sample uses expo-image-picker:

import * as ImagePicker from 'expo-image-picker';

const { granted } = await ImagePicker.requestCameraPermissionsAsync();
const picked = granted
  ? await ImagePicker.launchCameraAsync({ base64: true, quality: 0.8 })
  : undefined;
const asset = picked && !picked.canceled ? picked.assets[0] : undefined;
if (asset?.base64) {
  const scan = await january.foodAnalysis.analyzePhoto({
    image: `data:${asset.mimeType ?? 'image/jpeg'};base64,${asset.base64}`,
  });
}

Keep photos to about 1,000 px on the longest edge, the size the iOS SDK's PhotoScanImage sends. Resize with an image library such as expo-image-manipulator before you encode the photo.

analyzePhoto uses the reasoning-based analyzer by default. Pass reasoningEffort: 'none' for the standard one; the result shape and cost are the same.

Log a detection

Each detection's food is ready to log:

Field	Meaning
food.serving.id	The catalog serving to log
food.serving.quantity, food.serving.unit	The size of one serving
food.quantity	Servings eaten; becomes FoodSelection.serving.quantity
food.nutrients	Already scaled to food.quantity

The API always returns food.id, food.serving.id, and food.quantity, but the TypeScript types mark them optional, so guard them:

const foods: FoodSelection[] = scan.detections.flatMap(({ food }) =>
  food.id && food.serving.id && food.quantity != null
    ? [{ id: food.id, serving: { id: food.serving.id, quantity: food.quantity } }]
    : []
);

Pass foods to foodLogs.create.

Correct a result

Send the complete analysis and an instruction:

const corrected = await january.foodAnalysis.correct({
  analysis: scan,
  instruction: 'The drink was unsweetened and there was only one slice of toast',
});