Can AI Clone a Real Person in One Hour? Inside Alibaba’s Virtual Human “Li Zhi”

Alibaba’s Machine Intelligence Technology Lab introduced “Li Zhi”, a highly realistic virtual human that can be generated from roughly one hour of video footage.

Presented at the CVPR 2018 conference in Salt Lake City, the avatar attracted attention from Apple, Facebook, Nvidia, Samsung and many researchers.

“Li Zhi” combines speech recognition, speech synthesis, machine Q&A and facial reshaping technologies, the latter co‑developed with Zhejiang University, delivering lifelike voice‑driven facial expressions, lip‑sync, blinking and smiles.

The system can understand various Chinese dialects, including Northeastern, Sichuan, Henan and Cantonese accents.

During the live demo, audience members interacted with the avatar and praised its realistic audio‑visual synchronization, noting the impressive results achieved with only an hour of video data.

Alibaba envisions applications across visual interaction scenarios such as airports, stations, customer service, retail guides, medical consultations, government services, smart homes, games and virtual reality.

In early 2019, Alibaba partnered with Tsinghua University on emotional computing research, aiming to give machines five senses and emotional awareness, with “Li Zhi” eventually possessing emotional capabilities and agency.

Science‑fiction author Liu Cixin remarked that such emotional‑computing breakthroughs represent a “wedge” into the core of human civilization.