Liu Yexi video screenshot.
"My name is Liu Yexi".
Recently, the new beauty makeup expert Liu Yexi debuted videos and videos of the Internet. The video special effects part are full of advanced sense, the virtual people are realistic and vivid, the hair texture and hand movements are almost the same as those of real people, and the interaction between virtual people and real people is also extremely smooth. The video of this beauty master who can catch monsters is released less than 30 hours ago, and the number of fans has soared to 1.3 million. As of November 23, the number of likes on the first video reached 3.366 million, and the number of fans had reached 5.36 million.
Xie Duosheng, the founder of Chuangyi Technology, the company behind Liu Yexi, told Beike Finance reporter that the two-minute video is only the preview of Liu Yexi's appearance, and the subsequent story will be released in Douyin in the form of a single episode. At present, Chuangyi Technology Company serves more than 150 large and middle-end teams behind Liu Yexi, and the number of small front-end teams is less than 10.
In 2007, when " Hatsune Miku " sang for the first time with an electronic synthesized sound and was called "Your Highness of the Princess" by Japanese Akihabara otakus, many people thought it was just a carnival in the second dimension. With " Luo Tianyi " entering the Li Jiaqi live broadcast room to bring goods and "Liu Yexi" attracting millions of fans in one day, virtual people have unknowingly entered the life of ordinary people.
This year, as the concept of meta-universe became popular, the virtual person, as one of its elements, was also pushed to the forefront. "In 2018, when we entered this track, many people didn't know what virtual people were, but this year it seemed that everyone understood, and many investors were also looking for related investment targets." Chen Yan, founder of the Next World Culture Company, told the Beijing News Beike Finance reporter.
Beike Finance reporter interviewed virtual person practitioners and learned that virtual persons are currently divided into three categories: hyperreal virtual persons, virtual idols and virtual person interaction products, and have realized commercial value in different fields. However, the high production costs of thousands to tens of thousands per second, the technical difficulties of real-time rendering of , and the AI technology that is difficult to overcome the human brain have also become bottlenecks in the development of virtual humans.
Siren is catching facial expressions. Photo provided by the interviewee
●Creating
has virtual idols that attract more than 10 million fans, "a few minutes of video costs tens of thousands of yuan"
holds the idol poster in his hand, lined up, shouting "Happy birthday" in unison...Around November 2, students from the University of Chinese Academy of Sciences , Shanghai Jiaotong University , and overseas Cambridge, New York University , upload birthday videos on B station. The fan support lineup is strong. As for the protagonist, it is not a real person in the traditional sense, but A-Soul's virtual idol "Jale Carol".
A-Soul is a virtual idol group launched by Lehua Entertainment in November 2020. At first, it was boycotted by native virtual anchor fans. However, shortly after the broadcast, its delicate modeling and the excellent business qualities of the "people in the middle" made many opponents "black to fans".
In a year, the A-Soul team has become the top virtual anchor of B station, with the number of fans of member " Jiaran " on B station reaching 1.26 million. A-Soul became popular and even led to many subculture "memes" such as "I really like you", "cute pinch", and "Take Me away".
Virtual idol group breaks the circle and has the shadow of the explosion of the industry. According to the figures released by B website Chen Rui, chairman of B website, in 2019, more than 32,000 virtual anchors on B website started broadcasting.
However, crowding in the market to test the waters does not mean that everyone can see the dawn of success. A set of public data shows that as of August 18, 2021, among the 3,472 virtual anchors who were relatively concerned on Bilibili, 1,827 people had a monthly revenue of 0 yuan, which means that more than half of them did not earn a penny.
"In fact, as long as you design a set of 3D models and purchase a set of motion capture devices, you can become a junior virtual idol." Liu Wen (pseudonym), an observer of the virtual idol industry, told Beike Finance reporter that the technology used by ordinary virtual game anchors is based on facial motion capture. In other words, as long as you put a 2D or 3D "skin", you can become a virtual anchor and can broadcast live like a live anchor.
However, Liu Wen said that both motion capture equipment and 3D modeling require costs, and the better the results, the higher the cost of equipment and models, which has led to many virtual anchors not making enough money.
Mars Culture founder Li Hao started his virtual idol business as early as 2017. Currently, its virtual image "Momojiang" has more than 18 million fans across the entire network. He told the reporter of Beike Finance that the current virtual idols have similar production processes. "First use modeling tools to create 3D models and iterate continuously, then use motion capture technology to drive the movement of the character model, and find a 'man in the middle'."
Li Hao told reporters that most virtual anchors currently use "man in the middle" and motion capture technology to broadcast live. "The 'Hatin Miku' and 'Luo Tianyi' were both used electronic synthesized sounds when they were first released, but since the birth of the second batch of virtual idols in Japan, the 'People in the Middle' began to be used in large quantities. This is because the virtual idols must be more like humans when speaking. At present, in terms of technology, electronic synthesized sounds have a relatively high degree of matching with real people, but when singing, there will be obvious technical barriers, and the inability to effectively deal with effects such as breathing and airflow sounds, which will make the audience feel that something is wrong, and the use of motion capture technology can effectively reduce the production cost."
In fact, virtual idol live broadcast is one order of magnitude more than live broadcasts. It is understood that facial motion capture and body motion capture are currently different technologies, so in extreme cases, when a virtual idol appears in the live broadcast room, his face and body must be carried by two people separately. In addition, technicians are required to synthesize the two motion capture animations, and then synthesize the audio and video with the voice recording of "The Man in the Middle", so that the live broadcast room effect can be presented that the audience sees.
In April last year, the gimmick of "Luo Tianyi" and Li Jiaqi's live broadcast in the same frame attracted a wave of attention. During the live broadcast, there was a "car overturn accident" that Li Jiaqi could hear "Luo Tianyi" but the audience could not hear.
Li Hao told reporters that behind virtual idols, a team often needs to support them. "Take 'Momo Jiang' as an example, there are ten people in the content team, including directors, scripts, motion capture personnel, voice actors, etc. In terms of technology, depending on the videos released, animator also needs to modify the model of 'Momo Jiang'. The cost of an ordinary short video may be around 6,000 yuan, and the cost of a few minutes of customized videos is tens of thousands of yuan."
If the technology and animation team of virtual idols can be replaced, the "person in the middle" is undoubtedly the soul of virtual idols. Le Element launched the virtual idol project " Fight Bar Song Girl in September 2018! 》, After more than two years of operation, the "people among the people" of the six singers "graduated" in February this year and released a farewell video on Bilibili. After that, when the operator announced the re-recruitment of "the person in the middle", many fans said in the comments: "If it is still an old man, it is not acceptable."
"We are deeply bound to the "the person in the middle". If the virtual idol has no "the person in the middle", it will take at least a month to stop updating, because even if a new voice actor is found to train voices, it is okay to chat, but singing is easy to be seen through." Li Hao said.
Digital astronaut and digital journalist "Xiaozheng". Photo provided by the interviewee
●Bottleneck
Behind Liu Yexi's money-burning popularity: high cost and high technology wall
As early as half a year ago, Chuangyi Technology smelled the vent of the metaverse and began to build Liu Yexi, a virtual human IP, which has been constantly polished from market positioning, character setting, character production, storyline creation, shooting execution, post-production and other aspects. Liu Yexi's eastern face, Chinese style makeup and the identity of a monster catcher are in line with the current national trend. At the same time, the use of fluorescent elements in the makeup, the special effects full of sci-fi, and the post-tone tone of the cyberpunk style cater to the preferences of young people in the Z era.
Regarding Liu Yexi's popularity, Xie Duosheng, founder of Chuangyi Technology, said that it was not surprising. The team also discussed this when reviewing the game: 50% of Liu Yexi's popularity is because of the popularity of the metaverse, 30% is because of its 2.5-dimensional setting and technical level, and 20% is the construction of video creativity and worldview.Currently, most virtual people on the market operate in the virtual idol model, and can be roughly divided into training, personality, and two-dimensional girl groups. The time and space where virtual people live are mostly two-dimensional or three-dimensional. Chuangyi Technology's positioning of Liu Yexi is 2.5 dimensions - the second dimension is pure CG, the third dimension is the real world, and the 2.5 dimension is an existence free from the two.
As of now, it is difficult to realize the subsequent virtual human IP monetization methods such as Liu Yexi and others in the short term. Whether the money can be spent alone can maintain its long-term operation is still a question mark. However, in Chuangyi's strategic layout, there are two main ways to monetize subsequent virtual people such as Liu Yexi - the traditional IP economy and the future business possibility of the meta-universe.
Beike Finance reporter noticed that most of the virtual idols currently conducting live broadcasts are mainly the second-dimensional style, while virtual idols such as "Liu Yexi" and "Ling" that are similar to real-person views are more important to be called "super realistic virtual people". Such virtual people are often not live broadcasts, but appear on social platforms such as Weibo , Douyin, and Xiaohongshu. They attract fans through their photos and videos like Internet celebrities and receive commercial endorsements.
"We do not do traditional 2D, nor do we touch the field of virtual anchors." Chen Yan, the producer of "Ling" and founder of Next World Culture Company, told Beike Finance reporter, "The virtual people we launched are mainly used in the field of pan-entertainment and brand fields, and most fans are groups that pay more attention to fashion life. If fans of the second-dimensional virtual idol can be compared with fans on B station, then the fans of super realistic virtual people are more similar to those of Xiaohongshu."
Beike Finance reporter sorted out that taking "Ling" as an example, the commercial advertisements they received are mostly similar to fashion stars, including luxury goods and beauty brands.
Compared with the second-dimensional virtual idol, the cost of making videos of hyperreal virtual people has also reached a higher level. In an interview with a reporter from Beike Finance, the "Liu Yexi" team said that in more than half a year before the launch of "Liu Yexi", investment in R&D costs, personnel costs, and technical costs "far exceed one million".
Chen Yan revealed to Beike Finance reporters that in order to cover costs, the company has made very strict product planning. Before each product is concluded, it will conduct five or six internal evaluations, including in which scenario each IP is used and what level is the attempt to create. "Taking the 'ling' as an example, we will refine it to disassemble the video into 15 seconds, 1 minute to 2 minutes, planning a major event in a quarter, etc. Otherwise, if the weekly update or daily update, the cost will not be covered at all."
In fact, the early realist virtual person in China can be traced back to the high-fidelity digital virtual person Siren jointly launched by NExT Studios and Epic in May 2018. During the research and development of this project, we can see the expensive side of the virtual human industry technology.
Tencent Interactive Entertainment Engineer David once used the word "hard" to describe the Siren project in the article "The Birth of Virtual Digital Human Siren". "The rendering time of a movie picture is often several hours long. Compared with the film industry, all our calculations must occur at that time. The virtual human program runs at 60 frames per second, and all calculations must be completed within a 16 millisecond period." In the end, the top team from four countries completed the project after overcoming the technical bottlenecks brought by software and hardware and other aspects.
Currently, most hyperrealistic virtual human projects are still difficult to achieve live broadcast in real time. "Now, real-time (virtual people) are basically stylized characters on the market. Realistic styles are usually videos made with offline CG processes. Some are just using Unreal Engine as the renderer . We have always adhered to the real-time + real-life line, because our goal is to make real-time digital people real-time and implement real-time interactive scenarios." said Ge Cheng, deputy director of the New Technology R&D Center of NExT Studios.
With the iteration of technology, the application scenarios of hyperrealistic virtual people will become more and more extensive.On June 20, Xinhua News Agency and Tencent jointly launched the digital astronaut and digital journalist "Xiao Zheng". This super realistic virtual person will undertake the "on-site reporting" mission of manned space engineering and planetary exploration engineering that is difficult for ordinary journalists to achieve. Ge Cheng told Beike Finance reporter that the NExT digital human team has always maintained within 20 people.
●Change
Will virtual human + AI become the "guide" of the metaverse?
With the development of technology, the boundaries of hyperrealistic virtual people, virtual idols and even intelligent interactive products have gradually blurred. Whether the field of virtual people can achieve technological "unification" in the future will bring a lot of imagination space to the market.
Tencent Interactive Entertainment stated that under the "black swan" events such as the epidemic, people are isolated from each other and will increasingly need to interact and connect. Virtual people and the virtual world are not just entertainment scenarios, but also sociality and dependence between humans need to be considered. Digital people can play a greater social value. In addition to the identities of digital astronauts and digital journalists, "Xiao Zheng" will also have more interactions for users and young people in the future. After "Xiao Zheng" is known and loved by more and more friends, it can also become one of the "virtual idols" of contemporary young people who represent mainstream values.
"In fact, it is more important to have continuous communication with users than the beautiful appearance of a virtual person." Chen Yan said that his vision is to pursue the intelligence and scenario of interaction with users and virtual IP while maintaining the existing business line, and to develop into a "virtual life ecology" company.
"At present, Next World Culture is cooperating with top AI companies such as ' Xiaobing ', trying to expand more intelligent virtual products. In other words, virtual people are IPs, but they will also add many AI intelligent functions to meet the needs of various segmented scenarios." He said.
The development of technology such as character modeling, real-time rendering, voice recognition, and action recognition has allowed many practitioners to see the future application prospects of virtual people. At the opening ceremony of 2021 World Artificial Intelligence Conference , four virtual people and real hosts appeared on the same stage - B station virtual idol "Ling Yuan", Baidu "Xiaodu", Xiaomi " Xiaoai Classmate " and Microsoft "Xiaobing". Among them, the last three virtual people already have their own application scenarios, such as users can order songs on Xiaoai speakers by shouting "Xiaoai Classmate".
Beike Finance reporter saw that many virtual human products have been promoted on the market, such as iFLYTEK , Xiangxin Technology and other companies have successively launched TO B virtual human products.
A virtual person product sales business staff told Beike Finance reporter that using the finished virtual news anchor provided by it will cost about 1 million yuan a year. virtual anchor can automatically generate appropriate voice and expressions based on the input text content, thereby converting text reports into video reports. "If you customize your image, it will take millions, because we need to conduct motion capture and algorithm analysis of the models you provide."
"Many people look at virtual people and only see superficial things, but I always think that virtual people are just a carrier, and what ultimately promotes the development of virtual people is the needs of people. In the real world, when a person's body has defects, by building his own virtual image, he can re-select (self-image) in the virtual world. Through AI technology, a virtual human IP should also establish a more stable relationship with humans." Chen Yan said. Mei Tao, vice president of
, JD Group, said in an interview with Beike Finance reporter that the AI technology of virtual digital people may have disruptive effects in the future. "Take JD 's own digital people as an example, there are 2D and 3D cartoon digital people, as well as real digital people. Digital people involve a wide range of technologies, including visual and voice recognition, as well as voice synthesis and dialogue, and graphics. In the future, we hope that digital people can truly complete some tasks, such as chatting with children, accompanying the elderly, citizen hotlines, smart customer service, etc. For this reason, we build digital people with our own characteristics based on the rich practice of intelligent customer service in JD e-commerce scenarios, and hope to form relatively mature standardized services in the next one to two years.As you can also see, there are many startups now working as digital people. After several years of development, it can be said that digital people technology and products are about to reach an explosive period. "
It is understood that at the 2018 SIGGRAPH Asia booth, the "Sierren" once showed samples driven by AI, but at that time, AI could only have one round of dialogue.
"Now AI can realize multiple rounds of dialogue, which is more intelligent, and AI drives digital people, and the future is promising. The integration of virtual and real is the general trend in the future. Digital people can gradually resonate with the real world and can exist as a part of society. Think about the Transformers, Saint Seiya, Doctor, Calabash we watched when we were young, all were virtual characters and became IPs that influenced our generation. As digital people flourish, we can foresee more virtual characters that will become bonded to this real society in the future. ‘The person in the middle’ is a topic that digital people cannot escape at present, but I believe that in the future, AI can better drive digital people in certain specific fields. "Ge Cheng said.
"If virtual idols can broadcast live, there must be a 'one in the mean'. Super realistic virtual portraits cannot be broadcast live, "soulless". Smart voice in mobile phones can interact, but the interaction level is still too low. "Liu Wen told the reporter of Beike Finance that if virtual idols, hyperreal virtual people and virtual assistants already in the market are integrated, this may be the future of virtual people leading to the metaverse.
Beijing News Beike Finance reporter Luo Yidan Li Menghan Editor Wang Jinyu Proofreader Jia Ning
