Supporters of Marcus Endicott’s Patreon can access weekly or monthly video consultations on this topic.
China's digital human industry has matured rapidly over recent years, propelled by an evolving policy landscape, fierce commercial competition, and a regulatory apparatus that has moved from broad AI governance toward sector-specific rules targeting virtual beings directly. Although no single national-level plan titled "digital human development promotion plan" has ever been issued by China's central government, a layered framework of municipal, provincial, and ministerial policies has steadily shaped the industry's trajectory. Beijing led the way in August 2022 with the Beijing Action Plan for Promoting Digital Human Industry Innovation and Development (2022–2025), the country's first dedicated digital human industry policy, which set a target for the city's digital human sector to exceed 50 billion yuan by 2025. The Ministry of Industry and Information Technology followed in September 2023 with the Three-Year Action Plan for Metaverse Industry Innovation (2023–2025), which incorporated digital humans as a component of broader metaverse development. Shanghai, for its part, released its own metaverse key technology plan in June 2023, designating digital human generation and driving technology as a priority focus area. By the end of 2024, twenty-one provincial-level units and forty municipal-level units had collectively issued roughly three hundred specialized policies touching on digital humans, reflecting how comprehensively local governments had embraced the sector.
The commercial ecosystem grew in parallel. iFlytek (科大讯飞), one of China's leading AI companies, developed a digital human platform offering lip synchronization and expressive body language capabilities, features that found natural application in customer service kiosks through its AI Virtual Interaction Machine product line. In October 2024, iFlytek released its Spark Super-Realistic Digital Human, claiming it was the industry's first system capable of semantically coherent lip-expression-action generation in a super-realistic format. Silicon Intelligence (硅基智能), founded in 2017 in Nanjing by Sima Huapeng, pursued a platform approach, building tools and APIs that allowed clients to create custom digital humans rather than marketing a single signature virtual personality. By the first half of 2023, the company had created digital doubles for more than half a million people and served over fifty thousand live broadcast rooms. Its co-founder Yan Bojun described the production time as essentially negligible, reduced to the time it takes to press a button and wait for loading. By 2024, Silicon Intelligence commanded 32.2 percent of China's digital human agent market, placing it first domestically and second globally according to Frost and Sullivan estimates. The company filed for a Hong Kong IPO in late 2025, aiming to become the first publicly listed digital human company, with revenue having grown from 223 million yuan in 2022 to 655 million yuan in 2024.
State media organizations were among the earliest and most visible adopters of digital human technology in China. Xinhua News Agency launched the world's first AI synthesized news anchor, Xin Xiaohao (新小浩), on November 7, 2018, at the World Internet Conference, modeled after journalist Qiu Hao and co-developed with Sogou. By 2020, that single AI anchor had produced over thirteen thousand news reports totaling more than thirty-five thousand minutes of content. Xinhua continued expanding its roster, debuting Xin Xiaomeng (新小萌) in March 2019 as the world's first female AI anchor for coverage of the Two Sessions, and Xin Xiaowei (新小微) in May 2020 as the world's first 3D AI synthesized anchor, incorporating muscle model technology for more lifelike facial expressions and body movements. These anchors evolved from seated broadcast formats to standing presentations with full body language, and were deployed extensively during major national political events. People's Daily, China News Service, and CCTV followed with their own AI anchors, including Guoguo (果果) in May 2019, Xin Mei (新妹) in September 2023 for coverage of the Asian Games, and Xiao C (小C) in 2020 respectively.
The 19th Asian Games, held in Hangzhou from September 23 to October 8, 2023, after a one-year pandemic-related postponement, served as a major showcase for digital human deployments. The most prominent was the Digital Torch Bearer, known as Nongchaoer (弄潮儿), through which over one hundred million participants from more than 130 countries created 3D digital avatars via an Alipay mini-program. At the opening ceremony, a giant digital human composed of these avatars co-lit the main torch alongside swimmer Wang Shun on a 185-by-20-meter IMAX-scale screen, marking the first-ever digital torch-lighting in Asian Games or Olympic history. The system was developed by Alipay and Ant Group (支付宝/蚂蚁集团) using a Web3D engine called Galacean, along with AI facial recognition, motion capture, and blockchain technology. China News Service launched its AI anchor Xin Mei on opening day to deliver daily broadcasts and medal highlights. Alibaba Cloud deployed a digital sign language interpreter called Xiaomo (小莫) primarily for the Asian Para Games held in late October, a project that had taken roughly two years to develop using a dataset of twenty-five thousand signs. The organizing committee, working with China Mobile, also launched the Asian Games Metaverse on August 10, 2023, the first metaverse platform created for a major international sporting event. Translation services at competition venues were handled by approximately two thousand iFlytek hardware devices supporting eighty-four languages, a hardware-based system rather than a digital-human-interface solution. The digital human technology at the Asian Games was driven primarily by Alibaba, Ant Group, and their affiliates, along with Shiyou Technology (世优科技) for motion capture at the opening and closing ceremonies, rather than by Tencent, whose role at the Games centered on esports titles and broadcasting rights.
JD.com carved out a significant position in commercial digital human applications through its Yanxi (言犀) digital human platform, powered by JD Cloud. In July 2023, JD launched the Yanxi Large Model with virtual anchor capabilities. By late 2024, Yanxi digital humans served more than nine thousand merchants and generated over 14 billion yuan in incremental sales. During the 618 shopping festival in 2023, digital human broadcast rooms increased roughly 400 percent compared to the prior Double 11 event, and personalized digital human deployments grew fivefold from 2022 to 2023. JD claimed to have reduced individual digital human production costs to double-digit yuan from tens of thousands, a reduction of more than 90 percent. Lenovo reported that its AI virtual anchor achieved 2.3 times the daily transaction amount of human anchors. In a spring 2024 test, digital humans sold 40 million yuan worth of goods during Chinese New Year off-peak hours, improving conversion rates by more than 30 percent. Ping An (中国平安), one of China's largest insurance companies, deployed AI video consultants on digital screens across approximately thirty-four branches beginning in August 2019, achieving a 75 percent AI replacement rate for certain customer service functions. The technology was screen-based rather than holographic.
China's regulatory framework for AI-generated content took shape through a sequence of increasingly specific instruments. The Provisions on the Administration of Deep Synthesis Internet Information Services, jointly issued by the Cyberspace Administration of China, MIIT, and the Ministry of Public Security, took effect on January 10, 2023. These provisions required service providers to add conspicuous labels to deep-synthesis content that could cause public confusion, covering AI text generation, voice synthesis and cloning, face generation and replacement, and immersive virtual scene generation. The rules prohibited using deep synthesis for impersonation and fraud, and required providers offering facial or voice biometric editing to obtain separate individual consent. The Interim Measures for the Management of Generative AI Services, issued by the CAC and six other agencies, took effect on August 15, 2023, extending labeling requirements to all generative AI services. In March 2025, the CAC, MIIT, MPS, and NRTA jointly published measures for labeling AI-generated synthetic content, effective September 1, 2025, accompanied by mandatory standard GB 45438-2025, requiring both visible labels and embedded metadata or watermarks for all AI-generated content.
On the technical standards front, MIIT-administered industry standards for digital humans were released in 2023. The YD/T 4393 series, covering virtual digital human indicator requirements and evaluation methods, was drafted by CAICT under the China Communications Standards Association framework, with co-drafting participants including Tencent, Baidu, NetEase, and SenseTime. The standards addressed appearance, voice, motion, interaction processing, and multimodal dimensions. CAICT conducted assessments under these standards, evaluating approximately eighteen companies in its second batch announced in January 2023, with Baidu receiving the only Outstanding Level certification. In January 2026, MIIT released a mandatory national standard for digital human identity within a metaverse classification framework, establishing a "one person, one code" system requiring all commercial or communicative digital humans to carry unique identity identifiers for traceability. In July 2024, a National AI Standardization Guide was issued jointly by MIIT, the CAC, NDRC, and SAMR, including a specific section on digital human standards covering appearance, motion generation, speech recognition and synthesis, natural language interaction, capability assessment, and multimedia rendering.
The most consequential regulatory development arrived on April 3, 2026, when the CAC published draft Administrative Measures for Digital Virtual Human Information Services for public comment. Structured in five chapters and twenty-seven articles, the draft regulation introduced mandatory continuous labeling of digital humans during service display, required separate consent before using sensitive personal information for modeling or image generation, banned the creation of identifiable digital humans without the depicted person's consent, prohibited virtual intimate relationships for minors and the inducement of excessive consumption, and mandated that providers monitor user emotions and levels of dependence, intervening in cases of extreme emotions or addictive behavior. Penalties ranged from 10,000 to 100,000 yuan for violations, rising to 100,000 to 200,000 yuan where citizen health or safety was endangered, with the CAC leading enforcement alongside more than ten participating agencies.
The market itself grew at a pace that consistently exceeded projections. According to iiMedia Research, the core digital human market expanded from 12.08 billion yuan in 2022 to 20.52 billion yuan in 2023, a year-on-year increase of roughly 70 percent, and further to 33.92 billion yuan in 2024, growing approximately 65 percent. Grand View Research projected a compound annual growth rate exceeding 53 percent for China's digital avatar market through 2030, and Frost and Sullivan valued the 2024 AI digital human agent market at 4.12 billion yuan, growing 85.3 percent year on year. The 2D digital human sub-segment alone grew 101.2 percent in 2024, reaching 2.89 billion yuan. By the end of 2024, China had approximately 1.359 million digital human-related enterprises, up 36.9 percent from the previous year, with over 413,000 new registrations in 2024 alone. More than half of surveyed Chinese enterprises reported having used digital human technology by that point.
A wave of open-source releases in 2024 and 2025 accelerated accessibility across the industry. Silicon Intelligence open-sourced HeyGem.ai in March 2025, a tool capable of generating a digital clone from one second of video and a single photograph in thirty seconds, then producing a sixty-second 4K video, which accumulated over thirteen thousand GitHub stars. Alibaba's DAMO Academy released EchoMimic V2 for half-body animation, Tencent and Zhejiang University jointly released Sonic, JD.com open-sourced JoyHallo for Mandarin-optimized digital human generation, and ByteDance released its LatentSync lip-sync framework. ByteDance also released OmniHuman-1 in February 2025, enabling single-image-to-full-body-video generation. SenseTime created a digital recreation of its late founder, Tang Xiao'ou, in March 2024, producing a realistic nine-minute speech from limited data through its Ruying (如影) AI 2.0 digital human platform. The cost revolution was dramatic: traditional digital humans had required hundreds of thousands of yuan to produce, but by 2024 and 2025, large language model-powered platforms brought costs down to what industry observers described as the hundred-yuan level.
These advances brought new social and ethical challenges. During the 2024 Qingming Festival, the business of AI "resurrection" of deceased individuals surged, with over 1,900 merchants appearing on e-commerce platforms offering services priced from a few yuan to roughly 20,000 yuan. Families of deceased celebrities including CoCo Lee, Godfrey Gao, and Qiao Renliang protested unauthorized AI recreations, invoking protections under Article 994 of the Civil Code for deceased personality rights. Major platforms responded with varying approaches to digital human content moderation: Tencent's Video Account prohibited non-real livestream content including digital human e-commerce in June 2024, Kuaishou announced opposition to low-quality AIGC livestream content, and Taobao allowed digital human livestreaming in August 2024 but required vendor certification. In Hong Kong, police dismantled multiple syndicates using AI face-swapping for romance-investment scams, arresting twenty-seven people in October 2024 and thirty-one in January 2025, with losses exceeding 34 million Hong Kong dollars in one operation and 46 million US dollars in another.
The institutional infrastructure supporting the industry continued to formalize. The First China Digital Human Conference, organized by the China Internet Society, was held September 23 and 24, 2024, in Beijing's Zhongguancun district, releasing three landmark reports. The society's president, Shang Bing, delivered opening remarks as a digital human avatar. The 2024 China Digital Human Development Report issued at the conference declared generative AI to be the key driver of digital human industry innovation, fundamentally changing traditional production and operation modes. By the end of 2024, China had formulated more than thirty national standards and led the creation of nearly ten international standards in virtual digital humans and virtual reality. In the 2025 Digital Human City Development Index, Shenzhen ranked first, followed by Beijing, Guangzhou, Shanghai, Hangzhou, and Wuxi. China's internet user base, which stood at 1.092 billion as of December 2023 according to CNNIC data and grew to 1.123 billion by June 2025, provided an unrivaled domestic market for these technologies. Underlying hardware development, while serving AI broadly rather than digital humans exclusively, continued to advance, with Huawei's Ascend 910 series chips serving as a de facto standard for Chinese AI inference, Cambricon's Siyuan 590 entering volume production in the third quarter of 2024, and China Unicom and Huawei deploying 5.5G pilot networks in Beijing in January 2024 achieving ten gigabit-per-second downlink speeds capable of supporting real-time extended reality applications. The emerging categories of digital human deployment settled into a rough distribution of media humans accounting for about half of use cases, service humans roughly 30 percent, and industry-specific humans the remaining 20 percent, a taxonomy that reflected both the technology's origins in broadcasting and its expanding reach into commerce, government, education, and cultural tourism.
[Apr 2026]