In contemporary China, there is a saying that “at the end of the universe, it is live selling”. It basically says that every famous person will eventually start being a live commerce host. 

Xiaomi’s Lei Jun, New Oriental’s Yu Minhong, and many other celebrities have all started live selling. 

Tuesday this week (16 Apri), JD’s founder Richard Liu also did his first live session on JD’s JD Mall platform. In less than 60 minutes, he showcased 13 products, mostly food items including eggs, milk, steak, etc.

In an age where AI-based virtual/digital human hosts are prevalent in China, JD had specifically mentioned that what you saw was not Liu himself, but his digital human. 

The question is – could you tell? 


Hint: hand gestures. 

We showed the two clips to some members of Momentum Works community, most people said that they thought it was real Richard Liu, not a digital human. “I would not have not noticed if you did not point it out for me,” was the common answer. 

In addition to the hand gestures repeating themselves in the exact same way a few times, the rendering of the goods lined up in front of Richard’s digital twin is also not 100% realistic.

To create digital humans, a recording of the real human is needed to train the system. In this case, Richard Liu will need to sit in front of the camera and talk/gesture.  The system will then use his voice, tone, facial impressions and gestures together with an script to create the actual live video. 

Digital humans are already quite common in China’s live commerce sector, and a number of different models have evolved to help MCNs and brands increase efficiency, reduce cost, and quite importantly, prolong the hours of appearances of famous hosts. 

Some of the cases we saw during our Live Commerce Immersion to Guangzhou/Shenzhen were much more realistic and sophisticated. We will also visit a few leading AI/digital human companies (as well as MCNs that adopted them) in our upcoming Live Commerce Immersion (Hangzhou Edition) on May 28-30. 

Technologists we spoke to believe inference capabilities for digital humans will probably arrive within 2 years, judging from how fast OpenAI and other leading firms are evolving their models. 

When that happens, a tough question will be how much brands will want to cede inference powers to non humans, and how to hold them accountable. But until that, cost and efficiency are still the biggest drivers for digital human adoption. 

Thanks for reading The Low Down (TLD), the blog by the team at Momentum Works. Got a different perspective or have a burning opinion to share? Let us know at [email protected].


Previous articleE74 – The Impulso Podcast – Transforming digital transformation: lessons from world’s best digital bank
Next articleHow did a Sequoia-backed founder become a Buddhist monk
Jianggan Li is the Founder & CEO of Momentum Works. Prior to founding Momentum Works, he co-founded Easy Taxi in Asia, and served as Managing Director of Foodpanda. The two years running Rocket Internet companies has given him a lifetime experience on supersonic implementation, and good camaraderie with entrepreneurs across the developing world. He holds a MBA from INSEAD (GMAT 770) and a degree in Computer Engineering from Nanyang Technological University. Unfortunately he never wrote a single line of code professionally - but in his first job he was in media, travelling extensively across Asia & Europe, speaking with Ministers & (occasionally) Prime Ministers. Apart from English and his native Mandarin, he is also fluent in French and conversational in Cantonese & Spanish. He tried to learn Latin (for three years) and Sanskrit (for six months) as well. In his (scarce) free time, he reads, travels, hikes and dives. Pyongyang, Tehran & Chisinau are among the interesting cities he has been to.