Image source@Visual China
文 | 财经无忌，作者 | 白嘉嘉
Recently, a New Year's greeting video swept the Internet.
In the video, Musk, Bill Gates, Beckham and other overseas celebrities congratulate everyone on a happy New Year in a straight and round Chinese, and even use "high-end" expression skills such as "smooth sailing, the best of both worlds, and Sanyang Kaitai".
In the comment area, in addition to the praise of "Mandarin is really good", some sharp-eyed people recognized that this group of videos was actually generated through AI technology.
As a branch of AI, digital humans can be called one of the hottest business stories in 2023. As long as it costs a few thousand yuan, and it takes some time to record audio and video, the large model can train a digital doppelganger, from the shape of the mouth, the rhythm of speech to the body language, almost exactly the same as the person.
However, this blue ocean has recently made some waves - a digital human company called Heygen has been condemned by the industry's leading companies.
Heygen is its American name, and in China, it's called Shiyun Technology. The immediate benefit of having a company registered on both sides is that Heygen can operate in both markets without having to meet the corresponding compliance obligations.
For example, in order to prevent AI technology from being used in illegal activities, domestic companies need to identify customers when they provide services to customers, but Heygen has not set up a corresponding mechanism, which not only undermines the cost logic of the industry, but also makes this already young market more vulnerable.
One phenomenon is that targeting the traffic anxiety of small and medium-sized merchants, a large number of speculators use Heygen to produce digital humans, and claim to have the technology and ability to operate digital human live broadcasts, but after the actual broadcast, merchants receive endless violations and bans.
The problem behind this phenomenon is that the digital human industry chain is seriously fragmented, and some manufacturers who have mastered the technology deliberately allow the downstream to "grow savagely" in order to expand their market share and influence. Merchants are deceived by the excessive boasting of digital human capabilities by some service providers, which has become the price of vicious competition.
Digital Human Chaos: OEM, Shell, Piracy......
In July 2023, Sun Xu felt the chill of the catering market, and he frequently swiped the news of his peers closing stores or changing careers in the circle of friends, and the remaining part of his peers were violently launching various preferential activities in order to save themselves. "The elimination rate in this industry has always been high, but this year (2023) it is significantly higher," he said.
2023 can be called the "year of death" for the restaurant industry. According to Qichacha data, from January to September 2023, a total of 990,000 restaurants nationwide will be revoked, four times that of 2022. Among them, 180,500 were revoked in June alone, including some leading brands that once had a place in the industry.
Seeing that the market is getting more and more volatile, Sun Xu wants to find some new channels to attract traffic to the store. Live broadcast is his first choice, but he has stage fright as soon as he is on camera, and it is too expensive to recruit another anchor. Later, I tried to ask some local small Internet celebrities to visit the store, but the conversion rate was also worrying.
Perhaps because the algorithm "insighted" Sun Xu's intentions, several digital human videos were pushed to his mobile phone.
Although Taylor Swift's hit video hadn't appeared at the time, the finished products that were thrown were so realistic that it was almost impossible to distinguish them from real people. This kind of video often follows the same routine, asking the audience to guess who is the digital person among the next few people at the beginning of the film, and revealing that they are actually digital people at the end of the film.
If you go back to the roots, these videos are indeed made by digital human technology. However, the people who push these digital human videos to the "Sun Xu" may not be a company with the ability of the whole chain.
In order to lower the threshold for users, the head digital human company often provides a batch of public digital humans for users who are unwilling to clone their own image to choose, just like choosing a character in a game. At the same time, in order to make more people willing to try this new technology, companies often offer a certain amount of free time.
These "benefits" provide a large number of companies that do not have the underlying technology and operation and development capabilities to exploit loopholes. They directly customize videos on the official websites of leading companies as individuals, and use them to solicit business for their own companies, and even choose to directly transfer videos from other companies to their own drainage accounts.
In the process of looking for a digital human company, Sun Xu almost "picked his eyes", "198 yuan AI virtual anchor digital human", "17.6 yuan virtual anchor tutorial" and other advertisements abound. During this period, he also felt that something was wrong, and he always felt that some of the faces in the samples provided by the other party were very familiar, "Now that I think about it, I may have seen it on a short video."
In fact, not only users, but also head companies are quite troubled by the chaos of shelling and OEM in the market.
Founded in Nanjing, Silicon-based Intelligence is one of the top digital human manufacturers in China, and together with Fengping Intelligence in Beijing, it is known as the "South Murong, North Qiaofeng" of the digital human world.
Sima Huapeng, the founder of Silicon-based Intelligence, once said in an interview with Caijing Wuji that Silicon-based Intelligence's digital human videos are often stolen by others to attract customers. These companies do not have the ability to operate in the later stage and cannot help customers achieve better returns, but they are very good at attracting customers through marketing drainage and ultra-low prices that destroy the market.
In fact, the chaos of the digital human industry has attracted the attention of the society within a certain range. Fixed Focus, Self-Quadrant, AI Technology Review and many other media have carried out relevant reports.
Chaos is rampant because digital people are "too young"?
The reason why the digital human industry seems chaotic is essentially because it is still "young".
Many practitioners may not accept this view, if you count from the hand-drawn digital human, this technology has been developed for nearly 40 years.
But for a long time, digital human production can only be done through hand-drawing, CG, motion capture and other means, which is expensive and lacks the level of intelligence, and it is not so much a digital human as a soulless digital holster.
In 2018, thanks to the progress of deep learning Xi algorithms, the production cost of digital humans was reduced from 10,000 yuan and hundreds of thousands of yuan to 1,000 yuan, but in essence, it still solved the problem of the appearance of digital humans and did not form real productivity, so the digital human entrepreneurship tide in 2019 and 2020 quickly returned to calm.
It is not until 2023 that the large language model represented by ChatGPT brings the dawn of giving digital humans a "soul", and the industry has ushered in its real qualitative change. According to the "Digital Human Research Report 2.0" from Tsinghua University, the scale of the digital human industry is expected to exceed 100 billion in 2025.
The flip side of the promise is the lack of consensus in the market on digital humans.
At present, the concept of "digital human" is very broad, from virtual characters in animation works, to Microsoft Xiaoice, to Teresa Teng who is "resurrected" through holographic technology, and even to make photos "move the mouth", all of which are put into the pocket of digital people.
Unless consumers have conducted in-depth research on digital humans, they are easily confused by the seller's rhetoric of "digital humans can only be like this" and "this is digital humans", thinking that this is just another business gimmick.
The second problem with youth is that the market is still adapting to this new technology, and the rules are constantly changing.
In May 2023, Douyin released the "Douyin Platform Specification and Industry Initiative on AI-Generated Content (hereinafter referred to as the Digital Human Industry Initiative)", taking the lead in opening up AI-generated pictures, videos and digital human live broadcasts, but it also means that digital humans will be "managed".
As the easiest field to monetize, many companies are pinning their hopes on using digital humans to reduce costs and increase efficiency. But at the same time, with the increase in the number of digital human live broadcast rooms, there are more and more doubts, such as the effect is too fake, the market is chaotic, and the price is not transparent. People are beginning to care about whether to allow "super fertile" digital people to seize everyone's already fragmented time in the face of the explosion of information.
Although no other platforms have followed suit, legislation at the national level has never stopped, and new norms or regulations are being introduced almost every month.
The changing market rules test the upstream digital human suppliers, even if they have the intention to do business in a down-to-earth manner, it is difficult to do solid services in the case of limited team size. At present, the blocking of digital human live broadcast rooms and accounts is still one of the most reported problems by users.
Professionalism is the only way out for digital people
After being blocked 3 times, Sun Xu gradually realized that he seemed to have been cut leeks.
At first, he was very satisfied with the digital human staff, and once used it to sell group purchase coupons for stores for 13 hours in a row. Although the digital human turned a deaf ear to the audience's questions during the process, he believed that he had ridden the free ride of the new era, and it was only a matter of time before the digital human became more and more intelligent.
However, the problem soon appeared, because it was judged to be recorded, Sun Xu's live broadcast room ushered in its first ban.
Recording and broadcasting is one of the easiest pitfalls for digital human live broadcast at present, which only plays digital human videos in a loop according to the script recorded in advance, which is essentially different from digital humans who generate content in real time with a large model. These differences are difficult to discern with the naked eye, but they can be recognized at a glance in the data monitoring background.
"The first batch of businesses to try the digital human live broadcast technology, many of them were recorded and cut leeks. Zhang Xian, the person in charge of the digital human agency company, said that many businesses have not done live broadcasts before and are not familiar with the rules of the platform, but they are very interested in digital people, and spend thousands of dollars to buy a digital person who has recorded a video from the agent, but it is easy to be blocked."
In addition to recording and broadcasting, problems such as poor quality of digital humans, poor lip syncing, low interaction ability, and video quality may lead to a decline in the customer's user experience, or even banning. Although some companies provide guidance services and will teach customers how to unblock the live broadcast room step by step, on the whole, there is still no company that can guarantee that they will not be banned.
However, there are also some companies that have used digital humans to double their efficiency.
Silicon-based intelligence revealed that the digital human live broadcast room of a leading brand in the wine and travel industry has harvested more than 1 billion GMV (gross merchandise transaction value) in 2023, and the single digital human live broadcast room of a coffee brand has harvested 856,000 revenue in just 5 hours in a single day.
Why do some people make money with digital humans and others can't?
Essentially, a digital human is still a tool, not a subjective person, and its potential is highly correlated with the user's understanding of the industry. In other words, only those who know how to know how can use digital humans well.
Take the most basic interactions, for example. In order to prevent the illusion of large models from causing digital humans to talk nonsense in the live broadcast room, more than 90% of digital human companies in the market use "keyword matching" technology. Match questions and answers with keywords, and as long as the keyword is triggered, it can be matched and fed back to the audience.
Under the mechanical matching mechanism, the live broadcast room of digital humans is often boring. However, if the operator of the live broadcast room has a corresponding corpus and knows what the anchor says and how to arrange the rhythm of the live broadcast to mobilize the atmosphere of the live broadcast room, even if it is also based on the "keyword matching" technology, the live broadcast effect will show a large difference.
Objectively speaking, the threshold for digital human live broadcast is actually not low, it requires customers to have a certain understanding of technology, be able to distinguish the difference between recording and broadcasting and real digital people, and also require customers to have basic common sense about the live broadcast industry.
The index starts to break the upper limit of live broadcast, 24 hours a day...... These phrases are often used to describe the merits of digital humans. Compared with individual individuals, digital humans certainly have innate advantages in "reproductive ability" and "labor time". But if you put it in the entire live broadcast market, 24 hours a day and the explosion of the number of anchors have become a reality many years ago.
Perhaps, the live broadcast platform doesn't care whether the one in front of the screen is a digital person or a real person, the strict platform rules and recommendation mechanism are doomed, and only "professional" is the killer feature to stand out from the encirclement.