Virtual human B -end business, the new battlefield of the giants

The popularity of the Yuan universe has promoted the landing of virtual humanization.

Text ｜ qixin

Source ｜ 36 氪 Enterprise Service Comment (ID: qifudianping-36kr)

Cover Source ｜ Vision China

In 2022, virtual human be hot. As the "identity card" of the Yuan universe, everyone who enters the Yuan universe needs one or even more "ava".

According to the relevant reports released by the Internet, in the first half of 2022, the number of Chinese Yuan universe track financing reached 41, with a financing amount of 5.46 billion yuan. Among them, (virtual) digital people related financing from 21, accounting for 25.92%. Top capitals such as Sequoia Capital, SoftBank Vision, Gaoma Capital, Jiyuan Capital, Jingwei Venture Capital, as well as Internet manufacturers such as NetEase, Ali, Tencent, byte beating, all layouts and even tried the water in person.

(Picture: Financing information of virtual human head enterprises in 2022)

As a global digital economy, virtual human tracks have unique characteristics. On the one hand, it connects the C -side social application and has won many fans with a perfect appearance, which continues to be active in live broadcasts, brand advertising, variety shows, entertainment and other scenarios. For example, Luo Tianyi, Hatsune Miku, Liu Yexi, A-Soul, Xu Anyi. These "never collapsed houses" idols are becoming the new favorite of the brand, relying on the "content" core to create an IP business ecosystem.

On the other hand, due to the restrictions on technology and cost, the main buyers of virtual human beings are B -end enterprises. The core demand is to reduce costs and increase efficiency. Although the growth of such demand is not as strong as the C -end demand, they determine the upper limit and future of the virtual human track.

What is a virtual number?

Virtual digital people, also known as virtual human, digital people, number of wise men, etc., in a broad sense, a virtual figure with digital shape.

The Digital Working Committee of China ⼈⼯ Smart Industry Development Alliance and Zhongguancun Digital Intelligent Industry Alliance defines virtual digital people as: virtual ⼈ with digital shape. With the following three noodles characteristics: ⼀ is the appearance of ⼈, with specific appearance, gender, and personality characteristics; ⼆ is a ⼈ ⾏, with the ability to use words, facial expressions, and physical movements; It is the idea of ​​having a puppet, which has the ability to identify the external environment and can communicate with the ⼈.

Human beings can naturally read the information and emotions contained through language, expressions, and body movements, but virtual human beings can only identify external languages ​​through artificial intelligence technology and use algorithms to build "souls".

For example, in the dialogue scene with real people, through natural language understanding (NLP) and natural language generation (NLG) technology virtual human beings can identify text/voice information, through knowledge maps and deep learning technologies, natural responses that conform to human logic can be generated; You can also make an automated answer through keywords to match the knowledge base. Through voice synthesis (TTS) technology, the voice tone of simulated human beings will be restored to the facial expressions that coordinate with language through voice -driven Facial Animation (ADFA) technology.

In summary, AI technologies involved in virtual persons include: computer vision (CV), natural language understanding (NLP), natural language generation (NLG), automatic voice recognition (ASR), voice synthesis (TTS), voice -driven facial animation ( ADFA), machine learning (ML), deep learning (DL), knowledge map (KG), knowledge base (KB), AIGC (artificial intelligence automatic generating content), etc.

Category of virtual digital people

Virtual people are currently divided into two categories: identity virtual and service virtual person. The former representative is virtual idols, virtual anchors, etc., most of which have the shape of a real person, characterized by real -life drive, one -way output, and the business model with IP attributes.

(Photo Note: AESPA, a Yuan universe girl group launched by Korean entertainment giant SM)

The typical representative of service -oriented virtual human beings is digital employees, virtual customer service, AI assistants, etc., characterized by two -way multimodal interaction, computer -driven, and can interact with people in two -way. The principle of "drive" is based on deep learning such as AI+technology to train large -scale sample data to form a specific data model. Using algorithms and models, virtual people can identify and understand information like real people, and automatically decide how to feedback.

As the scope of sample learning expands, the "scope of business" of digital employees will also expand, but it will not exceed specific majors. For example, Cui Xiao Pan, the first digital employee of Vanke, is mainly the assistance of the financial department. He is responsible for urging various prepaid receivables/overdue invoices. Later, with the continuous accumulation of sample data, her duties expanded to the maintenance of social security provident fund information, but basically no basically did Jump out of the financial category.

There is also Sequoia China digital virtual employee Hóng, which uses deep neural network rendering technology to study small samples. After a weekly training cycle, its positioning is "investment analyst". You can read hundreds of business plans in one second and make information refinement and summary according to industry attributes and financing stages.

In general, the B -end virtual human business usually has three significant features:

There is a stronger demand for anthropomorphicization. For example, smart customer service, elderly/children's emotional companion robots, etc., can replace real people to a certain extent.

Vertical professional field. Because there is no technical conditions for the birth of strong artificial intelligence, deep learning is mainly sample learning in vertical professional fields. For example, AlphaGo defeated all human riders. AlphaGozero trained only 3 days and defeated AlphaGo, but all games were limited to Go. Product/service standardization and large -scale replication, thereby reducing the cost of corporate manpower operations. From this point of view, the C -end virtual idol/star needs to be customized because it needs to be customized.

Domestic virtual industry player ecology

In summary, there are currently three main types of virtual players in China: basic layers, platform layers, and application layers.

1. Basic layer player

Basic players are mainly motion/modeling/rendering/XR (extended reality, including VR/AR/MR) software and hardware manufacturers.

For example, the representatives of hardware manufacturers include: Nuo Yiteng, Ling Yunguang, XSENSE, Yingchuang Technology, Qingsuang vision, etc. The main business is hardware equipment such as motion/optical/VR/sensor. There are both entry -level equipment and professional high -definition equipment. The more expensive optical catcher equipment ranges from more than 100,000 to hundreds of thousands. The inertial catcher and the relative price of the equipment are lower, but the overall cost is also at a level of 10,000 yuan, and the equipment is relatively bulky. It requires a professional studio environment.

Because of the advancement of motion capture equipment and identification algorithms, the current modeling can already use low -cost computer vision catcher technology. This technology needs is more common. For example, the built -in camera of mobile phones/computers, and even a photo, a video can achieve modeling. Although the accuracy is really not high, it has greatly reduced the industry entry threshold.

Software manufacturers mainly include phase -core technology, mirror figures, Xugu futures, Global ink, cloud technology, and half -human cats. Such manufacturers mainly focus on the modeling/rendering links of virtual digital people. They have deep accumulation in computer vision and have the ability to quickly build virtual people. For example

In addition, there is a type of comprehensive manufacturer of partial content, which represents Shiyou Technology, Burning Mai Technology, Second World Culture, 8:88, AVAR, Shiyue Xingcheng, etc. These manufacturers are positioned as comprehensive digital human technical service providers to provide full -link virtual human products, but fundamentally, the core business of such players is "virtual human IP/digital asset operation", but it has it to have The dual city advantage of "technology+content", on the one hand connecting technical manufacturers of B -end, on the one hand, connecting the basic layer of software and hardware manufacturers, can also connect the C -end content market, provide extension services such as IP brokers Customer customization needs.

Such manufacturers generally have rich virtual IP assets, have strong content capabilities, and generally have their own star IP. For example, Dili Lengba/Tao Siman of the next -world culture, Ayayi, who ignites the culture, A Yang, Xiao Miao, and Xiaoai classmates of Shiyou Technology, Jiuli and Gao Yuanyuan of 8.8.

(Picture note: from left to right: 翎 ling, ayayi, Xiao Yang)

Second, platform layer players

The second category is platform players. There are three core players of such manufacturers: one is AI background manufacturers, the other is the Internet manufacturer, and the third is vertical ISV (independent software developer).

1.AI manufacturer

AI background manufacturers such as AI Four Little Dragons -Shang Tang, Kuangshi, Yun Cong and Yitu, as well as Microsoft Xiaobing, HKUST Xunfei AI virtual human, chasing one technology, and benchmark technology. The main core industries of such manufacturers are artificial intelligence technology and their applications. Virtual human business is only packaged and applied to underlying technology, not the main business.

Because the technical direction of AI manufacturers focuses on its own technical directions, the focus of virtual human business is different. For example, the "killer" applied by Shangtang's SenseMars Avatar is the core advantage of Shangtang in the field of visual AI/face recognition. You can use photos/videos to generate the virtual incarnation of the characters. HKUST Xunfei focuses on the virtual anchor scene, which is also an extension of its own AI intelligent voice technology advantage. The Microsoft Xiaobing model focuses on the direction of content recognition, emotional recognition and deep learning. It is a relatively common development framework. The business reflection is the production of artificial intelligence content, AI custody editor, X-EVA virtual human emotional companionship, etc.

2. Internet factory

Internet manufacturers are also important players on the platform layer. At present, large factories have their own layout in virtual human ecology. Baidu Smart Yunxi Ling, Netease Fuxi, volcanic engine, Hangzhou Li Weiye (exclusive investment), Tencent Yun Xiaowei Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro -Micro number, Alibaba Cloud (Dharma Institute XR Lab), Huawei Cloud (Metastudio digital content production line) Wait.

Taken together, the large factory test water is relatively cautious and usually has a closer relationship with its own business. For example, in the three major product lines of Netease Fuxi, in addition to the virtual human human beings, the other two major product AI anti -plug -in and AI competitive robots are applied to the game scene. In addition, NetEase also developed the immersive activity system "Yaotai", investing in the 3D virtual character social platform IMVU and the company's company Genies.

Byte beating in the virtual human layout, on the one hand, the virtual human image production tool comes with the volcanic engine, and on the other hand, the exclusive investment of the hardware is Li Weicha. Reality glasses, coupled with the social app "party island", the overall layout of the byte fully reflects the ambitions of the social socialization in the future. In the field of enterprise service, the four cloud service manufacturers of Ali, Huawei, Tencent, and Baidu have all launched virtual human products/solutions, but they are basically affiliated with the sub -module of the artificial intelligence product line, which is more like the past AI technology/products have been packaged and packaged to form new scene/industry solutions.

3. Vertical ISV manufacturer

The ISV (independent software developer) players of the virtual human vertical track are very heavy, and they are also the core players of the B -side. Representative companies such as Zhongke Shenzhi, Magic Enamel Technology, Magic Ren Intelligence, Black Mirror Technology, Mind Universe, Shadow Eye Technology, Shiyun Technology, Shenye Technology, etc.

Most of these manufacturers have one -stop full -link virtual human technology/products and service capabilities, but unlike comprehensive manufacturers mentioned earlier, ISV manufacturers mainly encapsulate virtual human capacity into standardized SaaS. It is provided to B -end enterprises in a lower cost and lighter way. There is no need to purchase bulky hardware and expensive driver software, nor does it require professional talents, that is, use it, it is efficient and convenient.

Taken together, such manufacturers mainly focus on the two major business directions: one is the generation of virtual human image with fast and batch, and the other is the synthesis of virtual content.

In direction, the platform generally provides a large number of preset templates, which allows enterprises to independently set parameters such as virtual human gender, image, sound, and quickly form a semi -personalized virtual image. ) Quickly form a 3D online virtual avatar.

Direction II, virtual content ability, uses TTSA technology (text -driven voice and animation technology), STA technology (voice -driven animation technology), transform the entered into this article into voice, and automatically synthesize audio and videos through virtual human image. In the later period Users can also adjust the audio and video details according to the timeline.

3. Application layer player

The third category is the application layer players, mainly from the top enterprises from film and television, media, games, finance, brands, government affairs, medical care and other industries. Such as Xinhuanet, Ali, Tencent, Netease, Pudong Development Bank, Everbright Bank, Ping An Bank, Douyin, B Station, Blue Optical Standards, Mango Super Media, Zhejiang Culture Interconnection, Huayang Lianzhong, Mihayou, Huaxizi, L'Oreal, L'Oreal, L'Oreal, L'Oreal, L'Oreal, KFC and so on.

Application layer players are rooted in their own business scenarios, and they are more in line with their own brand characteristics in technology and content. For example, in the banking scenario, virtual people mainly highlight the business attributes of customer service. On the entertainment platforms such as Douyin and B Station, the goal is to create a virtual human content matrix and explore the business landing ecosystem of the future Yuan universe. And Hua Xizi, L'Oreal and other beauty skin care brands have made virtual human beings into brand spokespersons, and continue to explore new business models in short video live broadcast scenarios.

Because of the particularity of the virtual market, B -side technology/product manufacturers and content manufacturers with C -end are inseparable. At the current initial stage, there is also a certain confusion in the domestic virtual digital market, which has led to serious homogeneous competition in the industry. This also brings some difficulties for user procurement.

At present, the overall trend of the industry is fusion. Whether it is software or hardware manufacturers, whether it is B -end or C -end manufacturers, it is transforming to integrated solutions/suppliers. For example, action capture equipment manufacturers generally match and sell software, AI provides solutions to live broadcast scenes, independent ISV self -developed AI algorithms, IP operators connect upstream and downstream manufacturers to create star IP.

Here, the 36 氪 Enterprise Services reviews the following ten virtual digital person head companies/software for reference for the company's selection.

1. Xiaobing Super Natural Virtual Human People

Based on the unique XNR (deep neural network rendering technology) technology of Xiaobing virtual human, Xiaobing virtual human besieged to accelerate the traditional rendering process through deep learning to extract features in a large amount of data. Users only need to enter voice or text to drive virtual human expressions, lip shapes, etc., perfectly engraving the expression, temperament, and image of the training target, making the visual effect of virtual human be more like real people. Xiaobing's digital twin production provides users with a complete solution that integrates data, virtual human customization, management, and live distribution.

2. Xunfei Open Platform

Xunfei Open Platform is the mobile Internet intelligent interactive platform launched by HKUST Xunfei, which provides developers for free: covering voice capacity enhanced SDK, one -stop human -machine intelligent voice interaction solutions, professional comprehensive mobile application analysis. Its product AI virtual anchor solution, with Xunfei's voice synthesis, face modeling, image -driven, image processing and other artificial intelligence technologies to achieve automatic output from text to video, and support the generating multi -language video generation.

3. Chasing one technology

Shenzhen Charian Technology Co., Ltd. is the leading artificial intelligence company and AI digital employee provider. We mainly attack deep learning and natural language processing, providing intelligent semantics, voice and vision AI full stack services. Our AI digital employee intelligent platform can be deeply integrated with business scenarios, providing different types of AI digital employees to meet the intelligent upgrade needs of various scenarios such as enterprises and government users, marketing, operation, and office, helping them reduce costs and improve efficiency , Improve user experience, drive innovation and growth. 4. Xiangxin Technology

Core Technology was founded in 2016. With the development vision of the "Yuan Universe" and "Create a more real digital world" as the corporate mission, focusing on the deep integration of computer graphics and artificial intelligence technology, promote XR technology innovation and industry Applications, independently developed "virtual digital person engines" and "ultra -realistic digital platforms" have obtained large -scale applications in more than a thousand domestic and foreign companies.

5. volcanic engine

The volcanic engine itself provides AI intelligent application SAAS products, including portrait human body, machine translation, content customization, machine learning and other applications. The platform also provides a virtual image production platform, which is the Web -side virtual image content production tool. It aims to combine visual, voice, and server rendering algorithms to achieve the content production ability of virtual actors based on virtual idol/cartoon image/digital person/live reconstruction.

6. Zhongke Shenzhi

Beijing Zhongke Shenzhi Technology Co., Ltd. was established in 2016. Its core team comes from universities such as Peking University and China University of Science and Technology. Mental movement generates a driver engine.

7. Magic Technology

Combined with self -developed or third -party intelligent dialogue systems and third -party engines to achieve the ability to build people, educate people, use people, provide one -stop capabilities to build AI virtual human products.

8. Matter Ren Intelligence

"AI Virtual Human SaaS Cloud Service Platform", Multi -Ren Intelligence is a new generation of artificial intelligence cloud service platform with intelligent virtual human as the core. Users do not need to have professional knowledge such as artificial intelligence and computer graphics. The SDK provided can easily have a lifelike, configured and upgraded digital character role in its own application, and then for smart cars, financial telecommunications, webcasting, advertising, tourism culture and creative, Yuan universe, etc. Various application scenarios provide high -quality smart virtual services.

9. The universe

MindOS uses the intelligent mentality framework of the universe self -research, which breaks through the restrictions of the traditional AL single -point ability, so that virtual people can not only communicate with human language, but also have visual, cognitive reasoning, their own memory and personality. Let the virtual human be truly the aboriginal people of the Yuan universe to accompany and serve every user.

10. Shiyou Technology

Shiyou Technology is providing customers with a "4U" experience with better technology, better products, better creativity, and better services. At the same time, through the technical experience and forward -looking vision that cooperate with international first -class manufacturers (including: Intel, NVIDIA, Dell, Matrox, etc.) in a direct or indirect way, Shiyou Technology will actively promote the research and promotion of fast animation technology and strive to put the core of the core Technology and products are deep and thorough.

- END -