The Death of the Digital Human
On October 16, 2026, Alibaba Cloud's official website quietly posted a product change announcement: the virtual digital human product has officially launched the delisting process. New purchases will be stopped on November 30, the renewal channel will be closed in May 2027, and by November 2027, this product once regarded as metaverse infrastructure will be fully shut down, with the console, work order channel and supporting technical support all going offline. There was no special press conference, no industry farewell forum, no media press release. A plain line of announcement text put an official end to a noisy industrial cycle.
Looking back exactly five years, in the late autumn of 2021, a virtual beauty blogger named Liuyexi released a two-minute-plus short video on Douyin. She gained more than a million followers overnight, the comment section was full of amazement, and brands, investors and entrepreneurial teams flocked to the scene following the traffic. Later, the team behind her reviewed this sudden viral success and gave a very clear breakdown: 50% of the dividend came from the trending metaverse concept, 30% relied on the technical capabilities of digital human rendering and motion capture, and the remaining 20% was the creative narrative of the short video itself. This review also laid an early fateful foreshadowing for the entire track: The rise and fall of virtual digital humans is not simply a matter of technical capability, their fate has always been tied to market trends from start to finish.
On one hand, major tech companies are delisting their products in an orderly manner and liquidating their businesses, archiving the "next-generation interactive carrier" that they once placed high hopes on; on the other hand, five years ago, the whole internet was in a carnival, business plans were flying everywhere, and everyone was convinced that digital humans would become a standard brand configuration, a new traffic entry, and the ultimate substitute for real people. In this huge time contrast, there are three consecutive "deaths" of this track.
This is not a sudden collapse, but a slow, layer-by-layer death along the three main lines of technology iteration, capital narrative and commercial implementation.
The First Death: The Good-looking Appearance Expires First
The first layer of value of virtual digital humans is technical spectacle.
Why did Liuyexi become popular? Because at that time, "looking like a real person" itself was a scarce capability. Ultra-realistic modeling requires top-level CG skills, motion capture equipment costs millions at every turn, and facial binding, hair solving, and subsurface scattering are all hard-core skills. In 2021, a virtual face that can withstand close-ups is backed by a long production pipeline. Audiences were willing to stay because they were witnessing a spectacle: an almost lifelike person living on the screen.
The scarcity of technology propped up the valuation of the entire industry. Brands rushed to claim virtual spokespersons, and the beauty, banking and FMCG industries all wanted to launch a perfect, flawless face. AYAYI made her debut on Xiaohongshu, cooperated with Guerlain, and became the first person in the metaverse; various ultra-realistic virtual humans were released intensively, and every face emphasized the same thing: we look more like real people than real people.
But spectacle has a natural shelf life, it has no real industrial foundation and audience, and loses its attractiveness once the trend passes. Virtual humans at this stage died under two forces.
The first is aesthetic fatigue. The biggest shortcoming of virtual humans has been hidden under their bright appearance since the day they were born: they have exquisite appearances but empty "souls". Most of them can only recite pre-written scripts, with restrained and stiff expressions, and their eyes lack the warmth of natural flow. The so-called interaction is nothing more than jumping between preset lines. The audience's amazement comes from their resemblance to real people, but after seeing dozens of similar, standardized beautiful faces, the surprise quickly turns into numbness. Real live streamers get tired, say wrong things, have impromptu touching moments, and are more authentic. In the former virtual live streaming rooms, the average audience retention duration kept dropping, and platforms gradually withdrew the tilted traffic support.
The second is the democratization of the technology itself. In 2023, AIGC exploded. Before that, ultra-realism was a craft that required years of accumulation; after that, generative tools turned the ability to "look like a real person" into something that can be bought for just a few dollars. After the threshold was flattened, the technology premium also approached zero.
Many founders of CG companies complained that their company's core asset, the ultra-realistic production process, was replaced by AI tools much faster than expected. In the past, customers paid for realism, but now they generate it themselves with software.
Technology iteration killed the myth of technology. When looking like a real person becomes a free and equalized capability, the first layer of value of virtual humans goes to zero. This is the first death of digital humans. There were no farewell announcements or update suspension statements, and brand virtual spokespersons were slowly removed from the homepage, no longer updated, and no longer appeared.
The Second Death: When Capital No Longer Believes in the Story
The second death happened on the books.
At the peak of the boom, there were two parallel sets of narratives in the industry. One was for brands and investors: virtual humans never have public scandals, can broadcast live 24/7, can be copied infinitely, and are the ultimate solution for low-cost marketing. The other was hidden in the bills of the studios: a high-quality short video costs hundreds of thousands, and the investment in an ultra-realistic image from modeling, binding to motion capture and rendering often reaches millions; subsequent scripts, operation, behind-the-scenes voice actors and content iteration are endless continuous expenses.
Capital was initially convinced because they believed in a bigger story: virtual humans are the entry to the metaverse. From 2021 to 2022, when the metaverse narrative was at its hottest, virtual digital humans were defined as the natives, spokespersons and first batch of residents leading to that virtual world. The story was big enough to temporarily ignore the commercial closed loop. Financing rounds came one after another, valuations rose steadily, and everyone acquiesced that they should burn money to occupy the market first, and the returns would be realized eventually.
But this story has a fatal flaw: it cannot generate revenue.
High-end customized virtual idols have very few clients, with high unit prices but long delivery cycles, making it difficult to scale; all possible monetization paths including advertising endorsements, live streaming sales, and virtual concerts have been tried, but none of them can cover the cost. A founder who had obtained financing said that their virtual idol took on more than a dozen brand collaborations a year, which sounded lively, but the labor cost of one offline concert was equivalent to the endorsement income of three quarters. Virtual human is an expensive, operation-heavy content business, and the vast majority of players do not have the ability to operate content at all.
In 2023, the two pillars broke at the same time. One was the shift of capital. ChatGPT ignited the frenzy of large models, and investors found a more attractive next-generation entry: no modeling, no motion capture, no behind-the-scenes operators needed, the large model itself is the interaction. Virtual digital humans were downgraded from the metaverse entry to an accessory of large models, and even ridiculed as the old era's PPT concept. After that, the financing heat of the virtual human track plummeted. An industry insider said that after the Series B round, he met with more than 60 institutions, but no one was willing to listen to the story anymore.
The other is the contraction of major tech companies. Pico laid off staff and adjusted its XR business, and Tencent and Alibaba successively withdrew from or shut down their virtual digital human related business lines. Even the companies with the most abundant capital were no longer willing to pay for a business with no clear return, and many start-ups disbanded one after another to find other ways out.
As capital receded, the bubble lost its support. Those start-up companies that relied on financing to survive collapsed: R&D teams were laid off, motion capture studios were left idle, and projects originally planned to create virtual idols in batches were put on hold. This time, the commercial fantasy of creating stars at low cost and getting the return quickly died.
Capital never buys technology, but the narrative of the next-generation entry. When the narrative goes bankrupt, the company goes bankrupt along with it.
The Third Death: The Dusk of Electronic Zombies
After the first two deaths, the track should have been cleaned up, but speculators took the last baton.
When the delicate idol route no longer worked, when major companies left and financing was hopeless, the survivors turned to a more brutal path: unattended digital human live streaming. Merchants bought template images, imported circular lines, hung up 24 hours a day without on-duty staff, no need to pay live streamers, and fantasized about picking up traffic through endless online time. For a while, the platforms were full of digital human live streaming rooms with repeated actions, mechanical tones, and the same segment of lines played in loops.
This is the most speculative stage of the entire industry. Many people entered the market just to find a way to sell goods lying down without investing manpower. The image on the screen repeats the product selling points over and over, the picture loops back and forth, there is no real response, no on-site communication, which is essentially a pre-recorded video played in a loop, covered with the coat of real-time live streaming. The price war followed: ultra-realistic customization fell from the million level to hundreds of thousands, 2D templates fell from tens of thousands to hundreds, dozens, and even low-cost monthly subscription packages, and profits were compressed to almost disappear.
The deterioration of the business model eventually backfired on the entire industry.
On the one hand, it is the collapse of reputation. When virtual humans are gradually equated with electronic puppets and money-grabbing tools in the public context, those companies that are seriously developing technology are also affected. On the other hand, supervision and platform rules are tightening. The regulations on marking deep synthetic content have been implemented, the Interim Measures for the Administration of Generative Artificial Intelligence Services have come into force, platforms have concentrated on cleaning up looped pre-recorded broadcasts and low-quality unattended live streaming rooms, followed by traffic restriction, weight reduction and account bans. Therefore, the cheapest and most convenient path has been blocked. Many small and medium-sized merchants left the market completely at this time: they spent thousands of yuan on the system, expecting non-stop orders day and night, but in the end only got very few views, extremely low conversion rates, and even their accounts were penalized. The so-called zero-manpower live streaming proved to be a gamble, and the entire industry's reputation lost the bet.
In short, speculators have drained the last value of virtual humans at the lowest cost, and personally buried the last reputation of the industry. This is the third death. After the noise fades away, only a small number of technical applications rooted in specific scenarios remain: customer service guidance, government news broadcasting, enterprise exhibition halls. They are quiet and restrained, no longer taking replacing real people and detonating traffic as their slogan, but silently serving as tools.
However, from a dialectical perspective, virtual digital human technology has not died out. In fact, it has made a comeback in another way: in the era of large models, everyone can generate their own digital avatar, and AI live streamers and digital employees have re-entered the enterprise procurement list. This time, no one promises no public scandals or no fatigue, it only acts as a tool, quietly staying in the position where it is truly adapted.
What really died is the once explosive industrial myth that promised to create wealth. It died from the shelf life of technical spectacle, the bankruptcy of capital narrative, and the dusk of speculators.
The three deaths have successively buried the technology premium, the patience of capital, and the reputation of the industry. Many years later, when new trends blow again, we may remember this lesson: the destruction of an industry often happens because too many people are eager to extract value from it.
This article is from the WeChat official account "Naoji Ti" (ID: unity007), written by Shanhu, published with authorization from 36Kr.