Digital people are "on fire" again.
On June 13, 360 officially released the "360 Intelligent Brain" large model application, and at the same time, 360's AI digital human was also officially launched. On the closed beta version of the 360 Wisdom Brain model, digital people including "Monkey King", "Zhuge Liang" and "Einstein" have been launched; It is understood that at present, 360 AI digital humans have more than 200 roles, and also support customization to generate their own exclusive digital humans. At the press conference, Zhou Hongyi, founder of 360 Group, created his own digital avatar and had a dialogue with digital people such as "Musk", "Monkey King" and "Zhuge Liang".
Zhou Hongyi said that 360's core innovation in large models is AI digital humans, and the most important application entry for future AI large models is digital humans, not browsers, search engines, etc.
As early as March this year, when Zhou Hongyi and Sohu founder Zhang Chaoyang talked about the then hot ChatGPT in "Dialogue Under the Stars", they mentioned digital people: "ChatGPT + AI digital human = Zhang Chaoyang's digital immortality, and enterprises that cannot take this bus may be eliminated." ”
Recently, a number of A-share listed companies with digital human-related businesses have also announced access to GPT-like large models through announcements, investor platforms and media.
Hot concepts always shift too quickly. From the meta-universe to the big model, but that is a year, the wind is rotating. The digital human has also become one of the concepts that have continuously stepped on the two major outlets of "meta-universe" and "big model" in recent years. In fact, after the popularity of the meta-universe gradually declined, the voice of digital humans also declined, and the problem of chaotic price system and homogenization of products appeared in the market.
From the entrance of the meta-universe to the present, digital humans seem to be gaining a "new life" with the help of ChatGPT and large models. But in addition to digital anchors, what is the value of digital people who are not yet familiar to the public?
Digital people, have occupied the live broadcast room?
In the Douyin account of business consultant Liu Run, it can be found that many of his recent short videos have been marked with the prompt "This video is made using digital human technology". Among them, none of them are real people, but Liu Run's digital doppelganger.
In the live broadcast rooms of many small and medium-sized businesses, digital people have long replaced real live broadcasts. These digital people broadcast product information and guide users to place orders and buy coupons.
A company said that they contacted a digital human provider, only need to submit a real person 3-minute speech video, the other party can quickly generate a digital human anchor, the image is similar to the real person, even the voice and intonation are very different. Although its actions are monotonous, it can only simply lip sync, shake its head, pierce its eyes, etc., and can only broadcast the text content given in advance, but for the live broadcast room that does not need interaction, it is still much more worry-free and labor-saving. Moreover, the cost of production is not expensive. Some manufacturers will directly provide customer templates to make digital humans, and the cost of using them is only two or three thousand yuan a year.
The virtual anchor is just a microcosm of the business track after the rise of the digital human industry.
In 2020, with the outbreak of the metaverse, countless virtual digital humans emerged. Including Winter Olympics weather anchor Feng Xiaoshu, Vanke employee Cui Xiaopan, Douyin beauty blogger Liu Yexi, virtual star AYAYI, and so on.
Among them, Liu Yexi released his first video in 2021 6 hours later, the account increased to 100,000 followers, and exceeded 1.3 million 30 hours later; A week later, Yoo Yexi's number of followers reached 4.3 million. At present, it has 8.42 million followers.
(Photo: Liu Yexi's work on Douyin account, screenshot of Douyin platform)
Before this, many people knew digital people, starting from virtual singers such as Hatsune Miku and Luo Tianyi, which were loved by young people in the "two-dimensional", and A-SOUL, a virtual idol group launched by Lehua Entertainment. But now, through short video platforms, countless ordinary consumers have been occupied by such highly anthropomorphic digital humans overnight.
With the development of computer CG, modeling, rendering and motion capture, speech synthesis, AR/VR and other technologies, the image of digital humans is also constantly iterating. From the initial two-dimensional 2D animation image, it has become more and more "difficult to distinguish between true and false". Like Liu Yexi, it is one of the representative images of hyper-realistic digital people, and the image is extremely realistic.
The development of the digital human industry on the "fast lane" is also the result of many factors.
Under the east wind of the metaverse, digital humans have been given greater expectations by the industry and broadened their imagination of application scenarios. For example, it can exist as another identity of the user himself in the metaverse space, as shown in the movie "Ready Player One", players can enter the virtual space through the device, bind their virtual "avatar" in it, and operate. This identity-based "digital twin", like WeChat and mobile phone numbers, will become an extension of the user's subject.
In the real world, the consumer Internet is close to the ceiling, and it is already difficult to dig further into the user's online time, the development of various mobile terminals, etc. In contrast, the virtual world is tantamount to a "pure land" to be developed. In the virtual scene, the digital human has naturally become a new entrance to interaction.
Inspired by this imaginative space, the concept of digital humans rises with the metaverse. At the same time, with the outbreak of content production demand such as short videos and live streaming e-commerce, a large number of digital human images and IP have emerged, and many companies have also taken the opportunity to co-brand with well-known digital people, or simply launched their own digital spokespersons. The digital human industry has entered a new era.
Early virtual singers, including virtual bloggers and virtual spokespersons, most of the producers' purpose was to create independent IP and further commercialize. This kind of digital human is also known as identity digital person, as a projection of the real person's image, it is set up with a unique appearance, personalized "persona" and so on.
Although these images are no better than real stars, without their own stories as support, it is inevitable that they are a little "ethereal", but for the new generation of "Generation Z" groups, this kind of play is not unfamiliar. Just as Disney's little fox image "Ling Na Belle" can become popular, this kind of digital human image has also established an emotional connection with consumers through personalities and has its own fans. By selling peripherals, providing image endorsements for enterprises, live streaming tips, and performing performances, they also bring revenue to their producers.
(Picture: Virtual human AYAYI, source: AYAYI Weibo avatar)
What has helped digital humans expand their application space is the development of AI technology. With the breakthrough of deep learning algorithms, natural language understanding NLP and machine vision and other technologies are becoming more and more mature, digital humans driven by "algorithms" have begun to appear, and digital humans who can interact intelligently are gradually revealed.
In the white paper "New Momentum of Enterprise AI Digital Human Digital Economy Development", the development of digital human is divided into five stages. Most digital humans, in the first two years, have reached the third stage - relying on algorithm-driven, digital humans can achieve lip shape, limbs, expressions and other actions, and it can interact with users in real time in simple dialogue scenarios.
This means that digital humans can "work independently" in some limited scenarios. With the launch of open source cross-modal deep learning models, breakthroughs in underlying technology, and the application of deep learning have also reshaped the production process of digital humans.
A large number of digital people began to pour into the live broadcast room, "seizing" the territory of some small anchors. Although its interactive ability is not so strong, it can also help many businesses control costs in terms of "quantity".
From this point of view, digital humans can exist as one of the "digital employees" of enterprises replacing real human services.
Another example is that in early 2023, Yu Liang, chairman of the board of directors of Vanke, posted on WeChat Moments, congratulating "Cui Xiaopan" on becoming an excellent digital employee of Vanke Group.
This is the second time Cui Xiaopan has been awarded. This Vanke employee with the image of a young woman is not a real person, but a digital person. After joining the company in 2021, through learning financial knowledge, Cui Xiaopan will choose the appropriate time to remind the receivable account while seeing the financial data, so as to achieve efficient turnover and cash management of social funds. Because the write-off rate of prepaid overdue documents urged by Cui Xiaopan reached 91.44%, Cui Xiaopan also won the Best Newcomer Award at Vanke Headquarters that year.
According to Yu Liang's article, in 2022, Cui Xiaopan's scope of work has expanded from capital and finance to investment, financing, engineering, cost, marketing, operation and other functions.
In addition, at the end of last year, China Merchants Group launched the digital employee "Zhao Xiaoying", which realized the landing of the China Merchants Accompanying Office Collaboration Platform, which can carry out work summary, information notification, etc., and has the ability of automatic business processing to assist employees to achieve the unification of data service entrance.
Can the "entrance" story of the metaverse still be told?
Interestingly, there is a clear difference between digital humans and other Internet products: although it became popular in the Internet era, it lacks the soil for C-end applications. So far, C-end users are only curious about it, and those who really put forward application requirements are B-end enterprises.
After all, B-end companies best understand the value of Gen Z users and are the most sensitive to the booming wave of technology. Application digital humans are also based on their understanding of new markets, for the purpose of acquiring new users, enhancing brand value, and enhancing user experience.
Based on IP, the value of digital people in brand marketing is easier to be tapped. Therefore, it can also be seen that most of the core scenarios of existing enterprise application digital humans are the use of virtual IP to create new marketing scenarios, attract new users, and also enhance the brand's sense of technology; In addition, through the intelligent interaction of digital humans and other functions, the service to users is realized. For example, virtual customer service in banks, virtual tour guides in tourist attractions, virtual teachers in online classrooms, etc., have become the most common digital people in the public.
However, the digital human industry that can explode based on B-side demand must also face such a test: it can stimulate the interest of users and can be lower than the original human employee application cost.
But digital humans clearly have their own limitations. Its application type, which is still dominated by identity digital people with IP, accounts for about ninety percent of the current market. With the rise of huge content generation demand in the live broadcast industry, its application scenarios are still mainly concentrated in the live broadcast field.
However, starting from live broadcasting, digital humans are also limited to live broadcasting, and application scenarios are still limited.
At the same time, limited by technology, the current digital human interaction ability is still limited, and the expression is relatively single. This is also the intuitive feeling of many companies and users who use it: many digital human anchors are limited to playing pre-recorded content; Or the conversation during the interaction is more awkward. Therefore, when bringing goods live, many e-commerce companies choose the virtual anchor + assistant model, or they are controlled by the "person in the middle" to put on the appearance of a virtual anchor for the real person.
In addition, the high cost of creating the current popular hyper-realistic digital human is also a major constraint. Liang Zikang, founder of Liu Yexi's producer, Chuangyi Technology, once said that Liu Yexi's production investment, including personnel costs and technical costs, is about one million yuan, and its first Liu Yexi short video launched in November 2021 costs about hundreds of thousands of yuan.
The current level of technology is not enough for digital humans to support the new "entrance" story told by the metaverse. Now, the hottest phase of the metaverse has passed; In the meta-universe, digital people have not yet appeared more convincing representative products.
When digital humans only have differences in appearance, but are similar in interaction and function, the problem of "homogenization" of such applications is also highlighted after the user's initial freshness gradually passes. Digital human providers have also begun to "roll" in price.
However, the imagination it expands has been combined with the real scene of enterprise applications. Therefore, when the digital human urgently needs to break the game, the new outlet of ChatGPT and large models gives it new opportunities.
The big model had a "timely rain"
At the recent press conference, 360 also said that digital humans will become the most important application entrance for AI large models in the future. Without the support of a large model, digital humans can only output content according to the established script, cannot communicate, and have no personality or memory. While 360 launched the digital human activated based on the large model, and proposed the concept of "digital person with soul".
For example, at the scene, Zhou Hongyi showed the "legal specialist" digital human color proofreading, which assisted the common official seal management and contract review problems in enterprises.
In fact, this is essentially the same as Cui Xiaopan, Zhao Xiaoying and other digital employees. But based on the big model, the "brain" of the digital human has been expanded.
"After the launch of the large model represented by GPT-4, the core change is that service-oriented virtual humans can do a better job of understanding user intent and answering questions." Wu Xuan, chief brand officer of Honnverse, once said.
Compared with the traditional dialogue engine, the digital human after accessing the GPT large model can effectively output replies closer to the human tone, and the understanding of many problems is more in place. For example, combined with the functions of AIGC, digital humans in the pan-entertainment field can achieve real-time interaction with audiences; In the fields of finance, cultural tourism, education, and medical care, it can better complete the communication work of intelligent customer service and exclusive consultants.
In this case, in fact, digital humans have been given greater industrial significance - it has become an industrial model to assist enterprises to carry out digital intelligence transformation, and the interactive entrance at the landing application layer. Similar to the current AI assistant, but it is more intuitive in vision and interaction than AI assistant, with a unique appearance and personality, which can be used as a brand marketing tool, enhance the brand value of the enterprise, and also narrow the distance with users; At the same time, it can also be connected with the information background of the enterprise and become a veritable digital asset of the enterprise.
In terms of interaction methods, with the breakthrough of AI technology, perhaps future digital humans will also be able to communicate with users more fluently through voice and other methods.
Therefore, Guosheng Securities has proposed that with the improvement of computing power and the upgrade of GPT-4 models, digital humans will become a killer application that carries multi-modal after GPT-4.
On the other hand, after the emergence of large models, the production process of digital humans has also been reshaped, and in the production process, digital humans have also achieved cost reduction and efficiency improvement, providing the possibility for commercial application.
After the popularity of the meta-universe concept has declined, the digital human track that has not yet found an application breakthrough in the field of live broadcasting is tantamount to ushering in another "timely rain".
On April 25 this year, Tencent Cloud released the intelligent small-sample digital sapiens production platform. Based on the general multimodal large model technology, only 3 minutes of live broadcast video, 100 sentences of voice material and other small samples are only needed to produce "digital sapiens" similar to real people through multi-modal data input and real-time modeling.
In April, SenseTime also launched the Ronin's digital human video generation platform under the "Daily Update" large model system, which can also generate digital humans from a 5-minute live video material.
In May, Xiaoice announced the launch of the "Human Clone Project", which requires individuals to collect data in as little as three minutes to create AI clones derived from their personality, skills, voice, and appearance. On June 1, Xiaoice first batch of Internet celebrity "clones" were launched, and users can communicate with AI clones in the X Eva App, including AI clones of Internet celebrity "Hanzo Forest", which caused heated discussions on the Internet.
(Picture: Xiaoice first batch of clones online, AI Xiaoice WeChat screenshot)
The digital human industry has once again entered an active period. In live broadcast rooms and industry events, it is not uncommon for different anchors, hosts and bigwigs to have digital "doppelgangers". Companies in vertical industries are also seeking new directions for the landing of digital humans. Like Shunwang Technology, it recently announced that its digital human "Xiaojing" has applied AIGC to the e-sports hotel industry to help hotel owners achieve intelligent operation, automatically generate marketing content, and meet the needs of merchants for digital marketing in physical operations. Singbar also based on the real image of its CEO Chen Hua to create an intelligent digital human "Tony" based on ChatGPT technology.
But it's still an industry in its infancy. The commercialization process of digital humans still needs further exploration. But in any case, ChatGPT is not the "opponent" of the meta-universe, but provides it with new impetus, helping digital humans create more interesting "souls" after AI technology provides them with a beautiful "skin bag".