China's Three Platforms and the Regulated Rise of Virtual Humans

It is tempting to read the emergence of virtual humans on China's largest content platforms as a story of technology outrunning its rules, of novelty arriving first and governance scrambling along behind. The three platforms that dominate the field — Douyin, Bilibili, and WeChat — invite that reading only until one looks closely, at which point the opposite becomes apparent. Each entered the virtual human field from a markedly different position, with its own architecture, audience, and commercial instinct, and each has been drawn toward the same destination: a settlement in which the digital human is not an autonomous curiosity but the visible face of an accountable, real-named, and increasingly licensed human standing behind it. What binds the three platforms is not a shared enthusiasm but a shared trajectory, and the force describing that trajectory is regulation. Virtual humans in China did not rise despite the rules imposed on them; they rose through those rules, which converted an unstable novelty into a durable, monetisable, mainstream presence precisely by binding it to the accountability structures already built for people.

Douyin sits at the permissive end of the spectrum, hosting the widest range of virtual human content of the three — from casual avatar tools available to any creator through to elaborately produced virtual idols backed by professional studios. The most visible of these, Liu Yexi, was assembled by a production company employing a team of more than 150 people and debuted in late 2021 to immediate virality, gathering over a million followers within a day and drawing hundreds of millions of views for her serialised mini-drama. Yet Douyin was also, revealingly, the first major Chinese content platform to publish its own governance framework for AI-generated content, issued in May 2023, well ahead of the national instruments that would later formalise the same principles. Its rules required that machine-generated content carry a visible watermark identifying the tool and the company behind it, that the real person behind any virtual human complete real-name identity verification, and that virtual livestreams be driven by a real human in real time — with fully AI-driven livestreaming explicitly forbidden and only 3D avatars, judged less prone to fully automated abuse, admitted to certification. The platform most hospitable to virtual figures was therefore also the one that insisted, earliest and most plainly, that a virtual figure is only ever a mask, and that the law's interest lies in the accountable person wearing it.

Bilibili approached the same problem from a wholly different origin. Founded as a niche community for anime, comics, and games, it grew into the natural home of China's virtual streamer culture, hosting the country's largest population of what are locally called virtual UP hosts, or Vup — a community that expanded from a few thousand at the end of the last decade to tens of thousands within a few years, and doubled again thereafter. Where Douyin's virtual idols were largely studio productions layered onto a mass-market platform, Bilibili's virtual streamers were grassroots performers native to the site's culture, and the platform built dedicated sections, technical standards, and monetisation channels around them, from tiered fan memberships to brand-collaboration marketplaces. What is telling is that this cultural centrality bought virtual streamers no special dispensation. Bilibili holds them to precisely the same conduct standards as human influencers, consistent with the national Code of Conduct for Online Streamers, which explicitly folds AI-synthesised virtual streamers into its scope. When several were found to have accepted tips from underage viewers, the platform issued forced refunds exactly as it would for a human host, and it has disciplined virtual performers over resemblance disputes in the same way. The virtual streamer, however beloved by the community, is answerable as a person would be.

WeChat represents the opposite pole, and in doing so exposes the logic the other two share. In June 2024 its Video Channels service banned digital human livestreaming outright, classifying both AI-driven and human-driven virtual avatars as non-authentic, low-quality content and excluding them from the tipping, e-commerce, and payment machinery that makes the platform commercially formidable. It was the first major Chinese platform to impose so categorical a restriction. The decision is striking because Tencent, WeChat's parent, is no stranger to virtual humans: it operates enterprise digital human services through its cloud division, has cultivated its own virtual idols and virtual news presenters, and has helped draft the very national technical standards for the technology that WeChat had just walled off from its dominant consumer surface. The apparent contradiction resolves once the underlying principle is seen. WeChat's objection was not to virtual humans as such but to inauthenticity — to a presentation that obscures who is really speaking and whether an interaction is genuine. That is the same concern that animates Douyin's real-person requirement and Bilibili's identical-treatment rule; WeChat simply enforced it by exclusion rather than by conditional admission, while aligning its remaining synthetic-content rules — prominent labelling, mandatory real-name verification — with the national framework.

Behind these three postures lies a national regulatory framework that has assembled itself with remarkable speed. In the space of four years a cascade of instruments took effect — governing algorithmic recommendation, deep synthesis, generative AI services, the labelling of AI-generated content and its companion technical standard, and livestream commerce — each adding a further layer of obligation around synthetic media and the figures built from it. What makes this framework so effective at producing convergence is its structure: it is a deliberately tripartite model, dividing the work of governance between the state, which sets the principles, the platforms, which are enlisted as front-line enforcers required to build the labelling, verification, and review systems into their own architecture, and the industry associations that codify professional conduct. The state does not so much police virtual humans directly as compel the platforms to police them, which is why three companies of such different temperament end up implementing near-identical mechanisms. The declared posture throughout has been one of inclusive and prudent supervision — tolerant of the technology's expansion, insistent on its terms. The trajectory culminated, for the moment, in a draft measure circulated by the Cyberspace Administration in April 2026 directed specifically at digital virtual human services: the most targeted instrument the sector has yet faced. It defines the digital virtual human in law, requires a conspicuous label displayed continuously throughout any appearance, and demands independent consent for the use of a person's biometric data in modelling, with parental consent for younger minors and deletion triggered when consent is withdrawn. It layers in protections that ban virtual companions and intimate relationships for minors and content designed to induce excessive spending, prohibits using avatars to evade facial and voice authentication, forbids the impersonation of heroes and martyrs, and attaches financial penalties to breaches. The draft's ambition is not to suppress the technology — it explicitly encourages its spread across every field — but to ensure that wherever a virtual human appears, the accountable human and the honest disclosure appear with it.

The convergence is the point. Three platforms that could hardly differ more — one permissive and commercial, one communal and subcultural, one restrictive and cautious — have arrived at a common settlement, because the regulatory field they inhabit rewards the same behaviours regardless of where a platform begins: transparency about what is synthetic, real-name accountability for who stands behind it, and behavioural standards identical to those imposed on human counterparts. Enforcement has kept pace with the rules, through mass removals of unlabelled synthetic content across all the major platforms at once, and through qualification requirements now imposed on virtual figures who discuss regulated subjects such as medicine, law, education, and finance. The effect is to close the gap between the virtual and the human almost entirely. A digital human in China is no longer a novelty operating in a grey zone but a formally regulated entity, monetised through the same channels, held to the same conduct, and answerable through the same person as any human creator. That is why the rise and the regulation cannot be separated. The rules did not arrive to tame a phenomenon that had already flourished on its own; they were the condition under which it was permitted to flourish at all, and the reason virtual humans in China now occupy the mainstream rather than its margins.