Meituan ZhiBo scales digital-human live streaming to 10,000 channels

On October 3, 2026, Meituan's technical team published details of Zhibo, its digital-human livestreaming product. According to the team, Zhibo is built around a complete technical loop for AI streamers, summed up in four goals: looking real, moving accurately, performing vividly, and selling well.

On the appearance side, the system supports 1:1 replication of a real person, and the team says a streamer's identity stays consistent even after changing outfits or swapping backgrounds. For motion and performance, Meituan claims the movements are smooth and natural rather than puppet-like, and that speech and action run together — the avatar talks and moves at the same time, with audio and picture kept in sync.

Underneath sits inference acceleration technology that the team says can support 10,000 concurrent livestreams. Meituan frames that capacity as what makes broad, low-cost adoption of digital-human livestreaming possible.

The post did not go into which models are involved, what parameters they use, what the setup costs, or when it becomes available. Those are the only details the published account covers.