Can Ant's AI Actually Count Calories in Chinese Food?

Ant Group's Ah Fu (阿福) assistant added an "AI snap food" feature on July 29, letting users photograph meals and receive automatic calorie estimates, then deposit the result into a personal "calorie account." The feature pairs an upgraded multimodal model with Bohe Health's roughly 1.6 million-item Chinese food database, accessed through Ant's prior strategic investment. The interaction runs in three steps: snap, log, edit.

The launch follows a 30-day-old "lose 100 million jin" campaign that has, by Ant's own count via its Tmall store, distributed more than 1 million body-fat scales and recorded nearly 1.5 million participants who claim to have lost more than 2 million jin (about 1,000 metric tons).

Regional Chinese dishes vary by hundreds of calories based on oil, sauce, cooking method, and portion density, and a single overhead photo only sees the surface. A richer label set does not resolve what the camera cannot capture.

Three empirical questions remain open: how recognition accuracy is measured across regional cuisines, whether the 1.5M participant and 2M+ jin totals are independently audited, and what Ah Fu adds beyond MyFitnessPal, Keep, Xiaomi, and Huawei's existing health stacks.