Liu Wei, one of the technical leads of Tencent Hunyuan, left to start his own business.

Written by Zhou Xinyu

Edited by Su Jianxun

"Intelligent Emergence" has learned from multiple independent sources that Liu Wei, one of the former technical leads of Tencent's Hunyuan large model, resigned recently.

As of the time of publication, Tencent has not responded to the above information.

Many practitioners told "Intelligent Emergence" that Liu Wei has started a business in Singapore, still targeting the field of video generation.

Liu Wei, the former head of the Computer Vision Center of Tencent AI Lab and the technical lead of the Hunyuan large model. Source: Tencent AI Lab WeChat Official Account

In Tencent, Liu Wei, who holds the title of "Outstanding Scientist", is best known as one of the technical leads of Tencent's Hunyuan large model.

Previously, he served as the director of Tencent Advertising Multimedia AI Center. Leading the team based on the Hunyuan large model, he developed an advertising multimedia AI technology matrix, improving Tencent's advertising conversion effect and advertising production efficiency.

Various signs indicate that before verifying the landing scenarios and business models, Tencent is still very cautious about the investment in video generation.

A practitioner said that images and audio are currently the focus of Tencent's multimodal research: "The technology is mature, the cost is controllable, and it is relatively easy to find scenarios." For example, in November 2024, the latest version of WeChat Official Account launched the "Reading Voice" function, using AI to achieve voice cloning.

"In Tencent, the main force of CV research is in the Hunyuan team. But overall, Tencent's investment in human resources and resources for video generation is not very large," the practitioner said.

Another relevant person said that the Tencent AI Lab Computer Vision Center has always been equipped with a small number of personnel, and Liu Wei took away several members after starting a business. "Overall, AI Lab is rather indifferent to video generation."

Regarding whether to invest heavily in video generation models, Chinese Internet companies have now been divided into two distinct camps:

The aggressive investment camp is represented by ByteDance and Kuaishou. In June 2024, Kuaishou released the highly effective "Keling AI"; in September of the same year, ByteDance, not willing to fall behind, released two video generation models at once: Seaweed and PixelDance.

The characteristics of these companies are that they have rich video data resources and business scenarios that can be immediately combined with video generation technology.

And the other camp that delays investment, such as Baidu, holds a more negative attitude towards the current commercial returns of video models.

Previously, "Intelligent Emergence" exclusively reported that at the director meeting in Q3 2024, Baidu's Chairman and CEO Li Yanhong clearly stated that the investment cycle of video generation like Sora is too long, and it may not be possible to obtain business benefits in 10 or 20 years. Therefore, no matter how popular it is, Baidu will not do it.

The attitude of Tencent towards video generation models is still not clear. Before clarifying the landing scenarios and commercial returns, the "Sora-like" video generation models will not become the area where Tencent makes a high-profile advance.

Welcome to communicate!