Supporters of Marcus Endicott’s Patreon can access weekly or monthly video consultations on this topic.
China's three dominant content platforms, Douyin, Bilibili, and WeChat, each occupy distinct positions in the country's rapidly evolving virtual human ecosystem, shaped by their respective platform architectures, user bases, and commercial imperatives. Taken together, they illustrate how digital humans have moved from novelty experiments to regulated, monetized presences operating under an increasingly formalized national governance framework.
Douyin supports multiple tiers of virtual human content, ranging from casual avatar tools accessible to everyday creators to elaborately produced virtual idols backed by professional studios. ByteDance offers several products for avatar creation and deployment: Douyin AI Avatar, a digital twin tool in beta for creators with over 500,000 followers accessible at shuziren.douyin.com; Xinghui, an AI camera application developed by ByteDance's Flow department; and Jichuang, a one-stop AI creative production platform with digital human video generation capabilities. At the enterprise level, Volcengine, ByteDance's cloud service platform, provides a full virtual digital human product line supporting 2D real-person, 3D cartoon, and 3D hyper-realistic avatar types. The Volcengine system accommodates both AI-driven and motion-capture-driven digital humans, with AI-driven subtypes including broadcasting, interactive, and perceptive modes. Its technical capabilities include lip-sync generation, facial expression synthesis, voice cloning from as little as ten minutes of audio, and digital clone creation from three minutes of video. A single T4 GPU supports ten concurrent 1080P streams at 25 frames per second, and the service is available through both public cloud API and private deployment. In February 2025, ByteDance advanced its avatar technology further with OmniHuman-1, an AI framework capable of generating realistic human videos from a single image and audio input.
Professional virtual content on Douyin relies on motion capture, facial tracking, and real-time rendering engines. The production company CREATEONE Tech, creator of the virtual idol Liu Yexi, employs motion capture with a team of over 150 people. Technical communities on Zhihu and CSDN document workflows combining motion capture with Unreal Engine 5, including MetaHuman character creation for Douyin virtual streaming. Third-party middleware such as ZEGO's RTC SDK enables real-time avatar streaming to the platform, while Douyin's own Live Companion software facilitates the streaming pipeline. Liu Yexi, who debuted on October 31, 2021, gained over three million likes and 1.3 million followers in under 30 hours, eventually accumulating between eight and thirteen million followers across platforms, with her mini-series attracting 380 million views. Xu Anyi, a top entertainment virtual streamer using a real-person motion capture model, reportedly earned over nine million yuan in gift donations in approximately 100 days of streaming. Despite these audience and revenue figures for individual streamers, the economics of high-end virtual idol production remain challenging. A UBS report estimated the average upfront cost of an advanced virtual idol at roughly 30 million yuan, and as of mid-2022, the company behind Liu Yexi had not yet turned a profit despite its massive following.
Douyin's monetization infrastructure for virtual content encompasses virtual gifting, e-commerce integration, brand sponsorships, and an affiliate marketplace called Jingxuan Lianmeng, a CPS affiliate platform connecting merchants and creators launched around 2020 and accessible at buyin.douyinec.com. Merchants set commission rates, and creators select and promote products through short videos or livestreams. Liu Yexi's team has received extensive brand cooperation requests, and virtual key opinion leaders such as AYAYI and Ling have conducted brand campaigns both on and off the platform.
Douyin moved early on platform governance for AI-generated content. On May 9, 2023, it published its Platform Regulations and Industry Initiative on AI-Generated Content, the first such regulation by a major Chinese content platform. The policy's eleven rules require that AI-generated content display an "AI-generated" watermark alongside the tool name and producing company in the upper-left corner, that the real person behind any virtual human complete real-name identity verification, and that virtual human livestreams be driven by a real person in real-time interaction, with fully AI-driven livestreaming explicitly prohibited. A follow-up governance announcement issued on March 27, 2024, addressed improper use of AI-generated virtual characters. As of 2023, Douyin accepts only 3D virtual humans for certification, excluding 2D and real-person clone types due to higher AI-only livestreaming abuse risk. Content moderation operates through a three-tier system of AI screening, human review, and recommendation audit. All uploads pass through machine learning models employing computer vision, audio fingerprint matching, and keyword analysis, with flagged content escalating to human reviewers for frame-by-frame assessment. Content reaching high view counts triggers additional review. ByteDance expanded its content review teams from 6,000 to over 10,000 people as of 2018, a figure that has likely grown substantially since. In March 2025, Douyin launched the Douyin Safety and Trust Center at 95152.douyin.com, publicly documenting its algorithm principles and review processes.
Bilibili occupies a fundamentally different position in the ecosystem. Founded in 2009 as a niche anime, comics, and games community, it has grown to 368 million monthly active users and 107 million daily active users as of Q1 2025, with professional user-generated video contributing roughly 89 to 91 percent of total playback volume. Bilibili hosts China's largest VTuber community, with VTuber numbers growing from approximately 6,000 in 2019 to over 32,000 in 2021 and doubling again by mid-2024. VTubers are known locally as "virtual UP hosts," commonly abbreviated as "Vup," and the platform launched a dedicated virtual streamer live streaming section on December 14, 2018, the first Chinese platform to do so, expanding it to a full section on February 16, 2022.
The platform's major industry collaboration in the virtual space is VirtuaReal, a partnership with ANYCOLOR Inc., the parent company of NIJISANJI, with recruitment announced on April 19, 2019. Bilibili led ANYCOLOR's Series A financing in August 2019 and remains its third-largest shareholder. ANYCOLOR went public on the Tokyo Stock Exchange on June 8, 2022, becoming the first VTuber company listed globally. VirtuaReal operates through three sub-projects: VirtuaReal Project as the main lineup, VirtuaReal Star for celebrity collaborations, and VirtuaReal Link for supporting independent VTubers. As of 2023, the project had 63 active members with total Bilibili followers exceeding nine million. A-SOUL, often mentioned alongside VirtuaReal, is a separate entity created by ByteDance subsidiary Chaoxiguangnian and Yuehua Entertainment, debuting in November 2020. A-SOUL streams on Bilibili but is not a Bilibili production.
Bilibili published 3D virtual streamer content standards on September 14, 2022, requiring motion capture technology or AI-based motion simulation, 3D virtual avatars as the primary streaming subject, and prohibiting looped pre-made animations, game characters, static images, or GIFs. In early 2022, the company acquired Shanghai Qianyu Network Technology, a firm specializing in 3D virtual idol live streaming software and facial motion capture synchronization. Bilibili's editing app Bicut includes built-in virtual avatar creation and face capture capabilities.
Bilibili's monetization channels for creators include the Charging Plan, a tipping system launched in 2016 and upgraded in 2023 with monthly subscription options; a tiered fan membership system for livestreams with Captain, Admiral, and Governor levels priced at 198, 1,998, and 19,998 yuan per month respectively; the Creative Incentive Plan for ad revenue sharing, restructured in 2024 with a 2,000 yuan monthly cap on basic incentives for smaller creators; the Huahuo brand collaboration platform officially launched on July 9, 2020; and the Bounty Plan, a program launched in September 2018 through which creators place CPM display or CPS product ads around videos with a revenue split of 50 percent for individual creators and 60 percent for MCN partners. VTubers dominate membership rankings, occupying all of the top 10 spots and 39 of the top 50 streamer membership positions, with 152,000 paying members generating approximately 3.5 million dollars monthly. The platform's Q1 2025 ad revenue reached 19.98 billion yuan, up 20 percent year-over-year.
Bilibili supports new creators through the New Star Plan, targeting those with fewer than 10,000 followers with cash prizes and special certification, and the Creator Academy, which provides online courses on filming, drawing, audio, editing, and special effects, supplemented by offline UP Host Academy Exchange Days in various cities. Bilibili World, the platform's annual offline event launched in 2017, regularly features virtual idol showcases. The 2021 edition at Shanghai National Exhibition Center, spanning over 100,000 square meters with more than 150,000 attendees, included a Virtual Paradise zone featuring Vsinger's Luo Tianyi and other virtual performers. The BML-VR concert series, a separate event for virtual idols, saw Hatsune Miku and Luo Tianyi perform together for the first time at the 2019 Shanghai Mercedes-Benz Arena concert before an audience of nearly 10,000.
Bilibili's moderation architecture mirrors the multi-layered approach seen across Chinese platforms. The Avalon system handles automated AI screening of comments, danmaku, and videos. Professional human review teams, significantly expanded after a 2018 government-mandated rectification with a new review center established in Wuhan, handle flagged and reported content. The Discipline Committee, a community volunteer body formally established in June 2017 with over 36,000 members by 2018, participates in crowd arbitration by voting on reported content. On August 29, 2025, Bilibili implemented the national AI labeling requirements through a Creative Declaration system requiring creators to self-declare the use of AI synthesis technology at upload, with dual-track labeling combining visible markers and metadata watermarks per the national standard GB 45438-2025. The platform holds VTubers to the same standards as human influencers, consistent with the national Code of Conduct for Online Streamers, which explicitly includes virtual streamers synthesized using AI technology within its scope. Enforcement actions have included disciplining VTubers for avatar resemblance issues and issuing forced refund notices to over 46 virtual streamers for accepting tips from underage viewers, identical treatment to what human streamers receive.
WeChat presents the most restrictive environment for digital humans among the three platforms. In June 2024, WeChat Video Channels banned digital human livestreaming outright, classifying both AI-driven and human-driven virtual avatars as non-authentic live streaming and low-quality content under revised operational rules. WeChat became the first major Chinese platform to explicitly restrict digital human livestreaming. This action stands in stark contrast to the broader Tencent ecosystem, which does offer digital human products through Tencent Cloud, including the Tencent Cloud AI Digital Human service supporting multiple avatar types with aPaaS interfaces and SDK access, and Tencent Zhiying at zenvideo.qq.com, which offers AI digital human broadcasting and video generation priced at 698 yuan per year for professional membership. However, these are enterprise cloud services deployed primarily through standalone applications, websites, and kiosks rather than deeply integrated into WeChat's core platform. The ban effectively prevents digital humans from accessing WeChat's otherwise robust monetization infrastructure, including Video Channels tipping, mini program e-commerce, and WeChat Pay integration.
Tencent's virtual character activities exist largely outside WeChat proper. Tencent Music Entertainment launched TMELAND, China's first virtual music festival, on December 31, 2021, in partnership with XVERSE, drawing 1.1 million participants across QQ Music, Kugou Music, QQ, WeChat, and Tencent Video. Xing Tong, Tencent's virtual idol originating from QQ Dance in 2017, has conducted brand partnerships with Levi's, Li-Ning, and Make Up For Ever, performed on CCTV and Jiangsu Satellite TV, and appeared on the 2026 Shenzhen Satellite TV Spring Festival Gala. Xiao Zheng, a digital journalist jointly created by Xinhua News Agency and Tencent, debuted in June 2021 for aerospace reporting. Tencent has also participated in drafting the national standard for Digital Virtual Human Technical Requirements issued by NRTA and co-drafted the Digital Human Financial Application Construction Guide alongside ICBC, CCB, Ping An Bank, and WeBank. WeChat's governance of synthetic content aligns with national requirements: Section 6.4 of the Video Channels operational rules requires prominent labeling of content created using deep learning, virtual reality, generative AI, or similar technologies, and real-name verification is mandatory for all livestreamers. On September 1, 2025, Tencent announced implementation of AI labeling rules across WeChat Public Platform, Tencent Music, and QQ Security Center.
All three platforms operate within a layered national regulatory framework that has developed rapidly since 2022. The Provisions on Algorithm Recommendation in Internet Information Services took effect on March 1, 2022, followed by the Provisions on Deep Synthesis in Internet Information Services on January 10, 2023, and the Interim Measures for the Management of Generative AI Services on August 15, 2023. The Measures for Labeling of AI-Generated Synthetic Content and the companion national standard GB 45438-2025 both took effect on September 1, 2025. The Supervision and Administration Measures for Livestream E-Commerce were released on January 7, 2026. Two draft regulations are currently under public review: the Interim Measures for Human-like Interactive AI Services, published December 27, 2025 with a comment deadline of January 25, 2026, and the Measures for Digital Virtual Human Information Services, published by the Cyberspace Administration of China on April 3, 2026 with a comment deadline of May 6, 2026.
The April 2026 draft regulation on digital virtual humans represents the most targeted regulatory instrument yet directed at the sector. Its 27 articles across five chapters draw legal authority from the Cybersecurity Law, Data Security Law, Personal Information Protection Law, and Online Data Security Management Regulations. The draft defines a digital virtual human as a virtual digital image using graphics, digital image processing, or artificial intelligence, driven by real humans or computers, simulating human appearance with speech, behavior, and interactive capabilities. It requires a conspicuous digital human label displayed continuously throughout any presentation, mandates independent consent for the use of personal biometric data in modeling with parental consent required for minors under 14, and provides that withdrawal of consent triggers deletion. Protections for minors prohibit addictive services, ban virtual relatives, companions, and intimate relationships for minors, and ban content inducing excessive spending. The draft prohibits evasion of facial or voice recognition identity systems, impersonation of heroes and martyrs, and false advertising. It requires that providers combine AI and big data technical measures with manual review, establish risk monitoring and early warning mechanisms, and ensure that government-facing digital human services include manual oversight and the right for users to reject digital human services. Penalties range from 10,000 to 100,000 yuan, with fines up to 200,000 yuan for harm to life or health. In a separate enforcement action in February 2026, the CAC ordered the removal of over 13,000 accounts and 543,000 pieces of content across Weibo, Douyin, Kuaishou, Bilibili, and WeChat for failure to label AI-generated material. As of October 2025, influencers discussing medicine, law, education, or finance must hold formal qualifications verified by their platforms. By December 2025, 748 generative AI services had completed CAC filing at the national level.
Academic analysis has described China's approach as a state-centric model of influencer governance involving the state, platforms, and industry associations. The Generative AI Interim Measures of 2023 explicitly adopted an inclusive and prudent approach with classified and graded supervision. The April 2026 digital virtual human draft encourages the application of digital virtual human services across all fields while establishing content safety requirements, and the December 2025 anthropomorphic AI draft states that the government adheres to combining healthy development and governance in accordance with law while carrying out tolerant and prudent regulation. The convergence of all three platforms toward transparency, real-name accountability, and regulatory alignment confirms that virtual humans in China now operate as formally regulated entities within mainstream content ecosystems, subject to the same behavioral standards, labeling obligations, and enforcement mechanisms as their human counterparts.
[Apr 2026]