Source |
Author丨Sakuragi Editor丨Ippah
The combination of bigwigs and digital people will never lack traffic.
Last week, the AI digital human "Purchasing and Marketing Dongge" based on Liu Qiangdong started bringing goods on JD.com's live broadcast, attracting 20 million views in less than 1 hour, and the final turnover exceeded 50 million. This achievement is also remarkable for the whole network.
The bigwigs who are at the forefront of the times and technology have no preference for digital people.
Three years ago, before ChatGPT broke out, Huang used a digital human to give a speech instead of himself. Because it was not disclosed in advance, no one could distinguish the real from the fake within three months. NVIDIA, whose main business is to produce GPUs that support the core hardware of digital humans, has since begun to soar to the sky.
According to the analysis of Guoyuan Securities, just as the current popular humanoid robot can replace laborers, the replacement of human beings by digital humans in the digital world, the reduction of costs, and flexibility and plasticity are all reasons that make it an outlet.
In recent years, with the improvement of basic hardware computing power and the upgrading of large model applications such as GPT-4, digital humans are expected to truly become personal clones, outputting text, images, audio and video, and even subtle emotional expressions. In the process of building digital content in the future, there will be more occasions where GPU + power consumption can be used to replace labor, truly creating a metaverse digital world, so that digital humans can integrate socialization, creation, and sharing, and become a multi-modal killer application.
So why has now become the best time for digital people to exert their strength?
First of all, from a technical point of view, with the improvement of computing power, the new concept of digital human is breaking the fixed cognition of "leather man". From the outside, through ultra-fine rendering, the high-resolution skin microstructure of the digital human greatly improves the realism of the digital human's appearance in the middle and close scenes, and the anthropomorphic expression is significantly improved, basically breaking through the previous criticism - the "uncanny valley effect".
Liu Qiangdong's digital human delivery from the C-end feedback, although there are many shortcomings, such as no interactive links, less portrait blinking, etc. However, with the investment of large manufacturers in computing power, it is believed that these problems will be gradually optimized in the future.
Second, digital humans are essentially a strong content-driven product. The 3D digital human that the market concerned about earlier had the disadvantages of "high cost, long cycle, and lack of interaction", which affected the commercial promotion. For a digital person with precision like Liu Yexi, the production cost is at least 500,000 yuan, and the creation cycle of each video takes about a month, and the behind-the-scenes creative team includes directors, planners, producers, 3D, operations and other personnel.
However, with the outbreak of vertical scenes, 2D simulation digital humans will expand rapidly in the direction of live broadcasting, entertainment anchors, customer service, game NPCs, etc., with their low technical threshold. According to the research report, the 2D simulated digital human is made using static scanning technology, that is, 40-60 cameras take all-round pictures of real people, and matrix scanning is carried out according to the light and angle of the photo, so as to present a 2D three-dimensional image in the software. Static scanning technology only needs to take a picture with a small amount of data required to create a 2D digital human image at a low cost. Especially with the rapid development of AIGC, the production threshold, cycle and cost of 2D simulated digital humans are far lower than those of 3D modeling digital humans.
(2D virtual news anchor, 3D modeling digital human)
Finally, on the content side, whether it is e-commerce platforms such as JD.com and Alibaba, or Douyin Kuaishou of short videos, they all have a huge precipitation of scene data. These data can support the whole process of e-commerce from multiple dimensions, from front-end sales, to mid-end customer service, and then to back-end after-sales logistics. With the development of AIGC, 2D digital humans also have language interaction functions, and because they are more vertical and easier to train with large models, the experience will continue to be improved.
In summary, cooling and efficiency enhancement + the improvement of content data have gradually opened up the potential of digital people in consumption scenarios. After the integration of online and offline new retail and interest-oriented content e-commerce, the blueprint of AI e-commerce has drawn a vague outline, and digital humans may be the key to open the door.