Editor's introduction: The players who have entered the virtual human track are gradually playing hot, and with the expansion of the market scale, the remaining problems in the development of the virtual human track have gradually emerged, and the technology is still difficult, and commercialization is relatively difficult... Specifically, how is the current virtual person market developing? This article has been interpreted accordingly, let's take a look at it together.
Unexpectedly, the capital feast of the beginning of 2022 was ignited by "them".
First, Hangzhou Li Weike Technology Co., Ltd. announced that it had completed tens of millions of yuan of angel round financing, which was exclusively invested by ByteDance; subsequently, Shiyou Technology, whose core technology was rapid animation technology and real-time digital human technology, completed the first phase of the 10 million-level A+ round of investment.
A month ago, virtual digital people and virtual fashion developer Shiyue Xingcheng completed the 10 million yuan angel round of financing led by NetEase, while the "virtual people" ecological company Nishi Culture has completed a total of three rounds of financing in the past year.
Jiyuan Mei, AYAYI, Liu Yexi, Ling _Ling, A-SOUL, Taosman, aespa, Hua Zhibing, Cui Xiaopan, Xi Jiajia... Virtual humans have entered our real world without a doubt. According to the data of the Head Leopard Research Institute, the current virtual digital human market size has exceeded 200 billion yuan, and this figure is expected to reach 270.3 billion yuan by 2030.
However, lively to lively, what progress has the development of virtual people made so far? Where is the commercialization value? What will the future hold? What other hidden worries are there? This series of problems may be the core of the attention behind the various virtual people.
First, the context
Virtual people are not generated overnight.
If you have to go back, I am afraid that you will have to recall the heroine of the 1982 Japanese animation "Time Fortress" Hayashi Myōmi. At that time, Lin Mingmei was packaged by the producer as a singer who sang an animated interlude and produced a music album, becoming the world's first virtual singer.
Since then, with the development of CG technology, motion capture technology, speech synthesis technology and other technologies, virtual people have continued to evolve - from the two-dimensional virtual Katsuhime Hatsune Mirai, the emergence of Luo Tianyi, to the current types of virtual digital people have been mass-produced, virtual people have made a significant leap forward:
- In terms of character design, more avatars are now hyper-realistic and high-fidelity, and they are not only very similar to real people in terms of image, but also closer to real people in behavior.
- Technological progress, whether it is the degree of intelligence of virtual people, or the smooth and natural degree of movement has risen sharply. According to the classification of technology, it is also divided into intelligent drive type and real person driven type (complete action through live action capture).
- From the perspective of practicality, virtual people have broadened feasible application scenarios, and there are highly instrumental service virtual people, such as virtual anchors and virtual employees, as well as IP virtual people who benchmark artists, idols, and Internet celebrities.
Li Shiyan, head of Baidu Intelligent Cloud AI Human-Computer Interaction Lab, believes that the digital human industry has three core driving forces, the first stage of rocket is user demand and technology upgrade, the second stage is policy support and capital influx, and the third stage is the iteration of the computing platform.
In the current situation, the three stage rockets are in a relatively early stage and are heating up. In the past few years, the value space of virtual people has not been opened, until the popularity of the concept of meta-universe last year, which pushed virtual people to the front of the stage and also made the complex industrial chain behind virtual people surface.
Gong Jun digital person
Tencent Byte Baidu and other big manufacturers, virtual people vertical track startups, technology providers, film and television production companies, MCN, IP planning and operation companies... They are all trying to take their place in this feast. There are many upstream and downstream links and roles in the virtual human industry chain, so the corresponding opportunities and challenges are also very different.
The "upstream" of virtual people is mainly the underlying technology providers of virtual people, including hardware manufacturers such as display devices, optics, sensors, chips, software manufacturers such as modeling software and rendering engines, specialized AI manufacturers, XR manufacturers, CG manufacturers, and vertical manufacturers of self-developed virtual human technology. Comprehensive Internet manufacturers such as Tencent and Baidu also have strong technical output capabilities.
The main player of "Midstream" is the IP planning and operation company, which completes the IP incubation, image design, operation brokerage and other work of the virtual person. Virtual human products are divided into IP and non-IP categories, the former is artist-type, the latter is service-oriented, paying more attention to practicality. There are many classification methods and dimensions, or divided according to application scenarios, or by technical ability, or by appearance. Guosheng Securities Research Report divides it into service-oriented virtual people (PGC+functional), virtual idols (PGC+IP value), digital avatars (UGC+functional), and creative carriers (UGC+IP value).
"Downstream" is the application of the scene, different application scenarios correspond to customers with different needs, covering the pan-entertainment industry, consumer goods and finance, cultural tourism, education, medical and other industries.
Combined with the "upstream, middle and downstream", it is not difficult to deduce the elements of creating a virtual person: the actual application scenario reverses the characteristics of the virtual person, and then coordinates the "content" and "technology" to land these needs.
However, at present, the nodes of the virtual person industry chain are relatively fragmented, and most of the companies in the industry are only involved in a certain link in the whole process of virtual people, so it is difficult to coordinate and optimize. At the same time, service-oriented virtual people and performing arts virtual people are independent, and it is difficult to make breakthroughs in personalization and scale, which also exposes the production efficiency of virtual people.
Second, create virtual people
The early stages of any technology are always accompanied by trial and error, iteration, and constant investment. The same is true of virtual people, a not so perfect virtual person, in fact, expensive.
For example, Liu Yexi, a "beauty expert who can catch demons", has released 6 videos since she was launched in October, and the number of Douyin fans is 8.345 million. According to Liang Zikang, CEO of Chuangyi Technology, the company behind it, "In 2018 and 2019, the price of our video per second was more expensive than 1 gram of gold, and now the price per second is equivalent to 2 to 3 grams of gold production level and production cost, and its core is to establish barriers." ”
Although he said in the interview that in fact, Liu Yexi's character creation cycle is not long, and it took less than two months to make the film, but the initial investment in huge research and development, personnel costs, behind-the-scenes content team of more than 100 people. Hardware and software equipment required for modeling, production, rendering, personnel teams... The virtual man can be said to be a gold-devouring beast.
Another virtual person industry practitioner told "True Detective" that Liu Yexi's team produced a film, "If there is no production fee of 1 million, it will not come out at all, if it is lower than this cost, the quality of the film will be complaining." ”
That is to say, hyper-realistic virtual human videos cost more than 10,000 yuan per second.
Dynamic Liu Yexi
Lin Yan, head of commercialization of Snow Claw Technology, a virtual person vertical manufacturer whose business covers virtual person planning and design, production integration and content operation, also expressed the huge investment. She told "True Detective" that although the company can currently achieve coverage of most links in the whole link, the initial investment is very large.
For example, snow claws to build a motion capture site of about 300m², need to be equipped with more than 40 limb capture equipment, as well as facial expression capture equipment, as well as motion capture suits, cameras, televisions, etc., the overall cost of more than 2 million yuan.
In terms of time, with the production process, the time period of the production of an avatar has been compressed, but from the original painting, model making, bone binding to adjustment, the overall process takes 6-8 weeks. "Each company's capacity for digital people is also limited." Lin Yan said.
Gina, a virtual person made by Snow Claw Palm Lab
According to the image, snow claws classify virtual people as hyper-realistic, 2.5-dimensional, 2-dimensional, Q version/cute pet, and the production cost of different precision virtual people is huge.
The entire production cycle of the Q version of the image from production to execution is only about one month, but a hyper-realistic virtual person takes longer.
Lin Yan told "True Detective" that she has a virtual idol that has not yet been listed, and the team has adjusted for nearly 8 months, "The time it takes to adjust and render the delicacy of her hair is the whole time of a two-dimensional image.".
Third, the future and hidden worries of virtual people
Although the production and operating costs of virtual people are high, as the application scenarios of virtual people become more and more extensive, the audience group continues to expand, and the enthusiasm of the demand side gives companies in the industry many opportunities to dig for gold.
In addition to IP-type virtual people, with the maturity of AI technologies such as NPL (natural language processing) and XNR (deep neural network rendering technology), more and more service virtual people have also found their own clear position in specific industries.
Recent cases include the daily economic news anchor and Vanke digital employee "Cui Xiaopan" announced by Xiaoice the company, the former can broadcast information uninterruptedly, and the face and form are no different from real people, and the latter can remind receivables/overdue and detect abnormal work under the blessing of the system algorithm.
Xiaoice virtual anchor of the company's cooperation with the Daily Economic News
Compared with IP virtual people, the technical conditions and application scenarios implemented by such virtual people are obviously different.
First of all, in terms of image generation, these two virtual people do not simply use expensive CG technology, but use XNR technology to extract features in large amounts of data through deep learning, speeding up the traditional rendering process. From the perspective of the audience's intuitive feelings, compared with the hyper-realistic virtual person, the virtual anchor image is more similar to the real person. In addition to image, the core competencies include reading, broadcasting, professional financial information writing, financial summary ability and so on.
Xu Yuanchun, COO of Xiaoice company, told Zhentan that in fact, the cooperation between Xiaoice and each one of them began in 2019, mainly focusing on financial information. At the end of 2019, the two sides had an in-depth exchange to think about what AI technology can do. At that time, the initial expectation was to make a "benchmark" Bloomberg financial TV, which could broadcast financial information around the clock. However, the technical conditions at that time could not be realized, people and voices needed to be natural and real, the integration of the entire system needed to complete real-time video push, and the video content interspersed in the middle should also be generated by technology.
It was not until the end of last year that the virtual anchor was officially launched. Behind it is Xiaoice accumulated technical investment over the years, and even difficult to quantify.
Baidu also released a virtual digital person Xiaopu in 2019, which cooperated with SPDB and the force animation of the virtual person technology company to apply it to the field of financial customer service. Last year, Tencent Games NexT Studios and the State Key Laboratory of Media Fusion Production Technology and Systems of Xinhua News Agency jointly created a digital astronaut xiaoyi to carry out news coverage for major Chinese aerospace projects such as manned space projects and planetary exploration projects. iFLYTEK launched an AI virtual anchor solution that combines speech recognition, semantic understanding, speech synthesis, avatar drive and other technologies.
Overall, service-oriented virtual people are mostly aimed at the needs of industrial digitalization, and their business models have certain similarities with SaaS.
The current charging model is to select the standardized ability module, "according to the logic of the employer to charge". Combined with the requirements of the scene, customers can select modules from the Xiaoice technology stack, form a package, and can adjust or upgrade the module selection according to the changes in needs.
"Now you can see that in some specific scenarios, the packages you choose are becoming more and more similar, which proves that this is the real use scenario." Xu Yuanchun said.
According to Xu Yuanchun, Xiaoice has the ability to complete the technology stack, in the natural speech understanding, computer speech, computer vision, artificial intelligence content generation and other aspects have been accumulated, providing a complete technical framework, with a variety of cooperation cases with industry benchmarking enterprises landing, the industry's professional cognition and technical reuse is also enhancing. "For example, now another TV station also wants to do a 24-hour financial TV, and the solution we have to solve is to retrain an anchor, and this process now only takes a week."
iFLYTEK's AI virtual human multimodal interaction service solution
The demand for entertaining virtual people is also exploding, and they have more "personality" and "personality" than service virtual people. But don't worry, the "human design" of the virtual person is completely controlled and will not collapse.
However, it is different from the precise requirements of the service-oriented virtual person for the professional quality behind the virtual person. The showbiz virtual person market is more "chaotic", the quality is uneven, the advantages and disadvantages are obvious, the technology and content cannot keep up with the concept, coupled with the high production costs, commercialization is not easy.
Due to the overlap of many technologies, many film and television special effects companies have entered the field of virtual people, such as the Hong Kong-listed visual effects production company Digital Kingdom. At the 2022 Jiangsu Satellite TV New Year's Eve concert, Digital Kingdom recreated "Teresa Teng" with high-end rendering technology, singing with singer Zhou Shen. Benefiting from the pull-up effect of the metaverse concept, Digital Domain shares rose for a week at the beginning of the year.
The film and television special effects industry essentially does customized services, and the bargaining power of the upstream in the industrial chain is weak, and the labor cost is high. The success of a blockbuster is usually inseparable from good visual effects, but the special effects production company is not on the list of box office dividends. It's not uncommon for top special effects companies to go bankrupt, such as the R&H behind "The Fantasy Drift of The Pi of Teenage PI."
For film and television special effects companies, cutting into the virtual person field is undoubtedly a good opportunity, but if this business model that does not dominate in the film and television industry does not change, I am afraid it will be difficult to profit from it.
Jiangsu Satellite TV New Year's Eve concert
Singer Zhou Shen performed on the same stage with the virtual "Teresa Teresa"
Another type of very active are vertical virtual person manufacturers and brokerage operation companies, such as Chuangyi Technology, Burning Wheat Technology, Second World Culture, Snow Claw Technology, etc.
The revenue of virtual IP mainly comes from performances, endorsements, live e-commerce, etc., according to the commercialization type, there are also two types: first, there are independent IP and then expand the scene; second, according to customer needs scene customization products.
Snow Claw currently with Tmall, Huaxizi, Kazilan, adopt a cow, Bailian, Vanke, etc. have carried out different ways of cooperation, according to Lin Yan, due to the difference in brand / platform customer needs, the way of cooperation will be different, there are long-term and short-term spokespersons, but the company promotes most of the cooperation of the annual frame, the magnitude of more than one million, in a year, based on the virtual person image, combined with marketing nodes, to provide service content.
The industry is on the rise, "compared to our previous investment, the profit is still minimal, but we are more optimistic about the future state," Lin Yan said. She believes that at present, it is still the stage of education brand and education market, many time, cost, use of problems, customers are not clear, but also their biggest concern before using virtual people, but with the more and more early brands, the benefit effect is obvious, she holds an optimistic attitude towards the industry.
There are many virtual human images in the market now, but the homogenization is serious, and creating a unique IP value requires strong operational capabilities. Head IP can make money, but the lack of a stable fan base, the personality is not distinct in the middle waist virtual person price is not high.
With a "god-level ID photo" in the little red book popular silver haired girl AYAYI, the first Po map received 100,000 + likes, but the average number of likes of the recent 10 new pictures was only 181; "expensive" Liu Yexi from october last year online Douyin, there are only 5 videos and 1 group of graphic content, content reserves are small, the company's follow-up launch of another virtual person is generally popular. Most virtual people are not highly recognizable, and they are oblivious and short-lived.
Blockbusters are hard to do, and harder to sustain.
AYAYI and Yoo Ye-hee's heat dropped
If the virtual person only stays in a perfect image, this is not sustainable after all, and more importantly, the collaborative integration of various technologies behind it, and its information can be conveyed as a content carrier.
In his dialogue with "True Detective", Xu Yuanchun was reluctant to define a role such as Vanke employee Cui Xiaopan as "service-oriented", hoping that the virtual people Xiaoice would be more humane, rather than just a link set up for the completion of the process work. "So we call them AI Beings, and we call them virtual people because it's a person in quotation marks." He said.
The virtual human track is still in its infancy, and with the development of the industry, the maturity of technology, and the reduction of production costs, the interaction between "them" and "us" may only be a normal thing in the future.
Author: Chen Wenqi, public number: True Detective AlphaSeeker
This article was originally published by @Deep Sound in Everyone is a Product Manager. Reproduction without permission is prohibited
The title image is from Unsplash, based on the CC0 protocol