The Digital Human Wealth Creation: A Storm Reshaped by AIGC
-
Talking about the company's plan to go public in the U.S. via SPAC next year, Zhan Jian couldn't hide his surreal excitement, 'I used to be an employee, never thought I'd become a shareholder one day.'
Three years ago, he joined the company, a live-streaming platform for e-commerce businesses headquartered in Hangzhou, the e-commerce capital of China, serving the Yangtze River Delta region. At its peak, the company received applications from 600 college graduates aspiring to be hosts, trained them for three months, and filtered out half of the paid trainees to select qualified hosts.
This was a high-cost experiment. The challenges became particularly evident during the pandemic when the company faced plummeting revenues due to travel restrictions and mass illnesses among hosts. After the 2023 Spring Festival, only 30 hosts returned to work. Unfortunately, shortly after their return, they received layoff notices. They were replaced by digital humans, which were gaining popularity in live-streaming rooms.
A decade-long AI researcher now working in the digital human field remarked that digital humans are one of the few clear-cut areas in AIGC: the technology is controllable—input leads to predictable output; computing power is manageable—generating facial expressions consumes limited resources, and there are vertical applications like short videos and live-streaming. In other words, it's a commercially viable path.
Zhan Jian's company is no longer the unlucky victim of the pandemic but a soaring success. In March this year, the company pivoted to become a digital human service provider, offering cloning and operational services for digital humans, along with tutorials on how to use these 'new weapons'—knowledge accumulated from years of running a live-streaming platform.
In the words of Zhan Jian and the company's founder, the company aims to be the 'Huawei of AIGC digital humans,' planning to go public in the U.S. via a SPAC merger next year and become a global digital human service provider. Zhan Jian is now one of the company's many partners.
Such ambitious plans are backed by the unexpectedly rapid growth in digital human subscriptions. In the five months since the pivot, they've cloned 1,000 digital humans, recruited agents in nearly 30 cities, generated over 30 million yuan in revenue, and secured tens of millions in funding.
Unexpected developments have been common in the digital human industry over the past six months: Silicon Minds, a pioneer in AIGC digital human development that began research three years ago, cloned only tens of thousands of digital humans last year but has now announced reaching 400,000 clones this year.
A marketing staff member at Silicon Minds recently expressed frustration: at the beginning of the year, the company set annual performance targets, which they achieved in just one quarter. The company then reset higher annual targets, only for them to be met again after another quarter, leading to even more ambitious goals now.
As one of the few profitable projects in the AIGC field, this sector has begun attracting numerous followers. An AIGC practitioner noticed that all five AIGC companies they interacted with had similar implementation plans: using "text-to-video" technology to generate digital humans for live-streaming sales.
Major tech companies have also entered the field: Tencent created Tencent Intelligence, Baidu developed the Intelligent Cloud Xiling Platform, JD.com launched Yanxi, and Huawei introduced the Huawei Cloud Pangu Digital Human Model.
Startups haven't been left behind either. Qichacha data shows nearly 20,000 new digital human-related companies established in the past six months. One industry insider confidently predicted that the sector would grow even hotter in the second half of the year, with growth exceeding at least 20%.
In this wave of AIGC transformation, the digital human field has achieved a small but significant goal: practical application and wealth creation.
Search for "digital human," and you're likely to see videos like this: a spacious office filled not with human employees (carbon-based lifeforms) but with rows of computer screens and non-stop digital humans (silicon-based lifeforms) broadcasting in live streams.
Being in such a scene doesn't feel more shocking than meeting an interesting person. But the changes it brings are evident - in the empty office corridors and workspaces, few employees can be seen, while digital avatars and display screens broadcasting live are everywhere.
Digital human technology companies and service providers all have such showrooms, which function similarly to a mobile phone store filled with demo units.
When meeting Zhan Jian in an empty room at 9 PM, there was no fatigue on his face as he eloquently discussed the company's development history and plans. Regarding going public, they plan to contact overseas institutions if next year's revenue reaches 20 million yuan. If not achieved, they'll have their already-listed investor acquire the company to complete stock conversion.
This represents a dreamlike reversal - from a company struggling during the pandemic to an AI startup planning an IPO immediately. Zhan Jian sees this undoubtedly as wealth creation: in the past, clients finding hosts on their platform would only generate 10,000-20,000 yuan per live stream, but now the focus is on expanding city agents, with each agent costing hundreds of thousands. Even selling digital avatars at 29,800 yuan/year is 'easier to sell than hosting services,' Zhan Jian said.
Metrics tied to revenue, such as digital humans and city agents, show clear data growth. One company achieved over 250% quarterly growth within five months of founding. At an AIGC sharing session, only Zhan Jian's company demonstrated tangible returns from applications—"4 months after launch, we collected 30 million yuan in payments."
A month later when reconnecting with Zhan Jian, they were already discussing their 4th round of funding and had raised revenue targets—having achieved the 20 million yuan goal ahead of schedule, the new target is 50 million yuan. The goal for the first half of next year is 30 million yuan.
The founder remains unfazed—this isn't the endpoint of wealth creation, but perhaps just a small signal. High revenue comes with significant investment. Last month alone, they spent 15 million yuan on chips and servers, scaling up to meet growing market demand.
"You can never achieve scale by relying solely on self-generated profits," he said of the company's strategy. The plan is to expand the market: "We must secure funding, go public, acquire competitors, establish monopoly, and go global."
Similar growth stories are emerging across digital human companies, reflecting wealth accumulation over the past year. One veteran digital human company founder revealed to media that AIGC digital human business in H1 2023 exceeded the total of the previous two years combined, with profitability already achieved.
Silicon Minds has set seemingly aggressive targets—to become China's first AI startup with 10 billion yuan revenue by 2024, and achieve 10 billion yuan profit by 2025. They plan to create 100 million silicon-based laborers including digital humans.
The business model explains these ambitions. After reaching certain database scale, digital humans can be generated at near-zero cost while quantities grow exponentially. Current prices for digital humans from some companies remain four times higher than a year ago.
The evolution of digital humans follows the direction of money.
Zhan Jian's digital human service company aims to participate in investment promotion conferences, recruit sub-agents, and sell individual orders while using small models to expand digital human applications in fields like medical aesthetics, legal consultation, and education.
Silicon Minds, a digital human technology company, sells its API interfaces and open-sources digital human cloning technology to business partners while venturing into publishing, film, and television. Their strategy is to collaborate with leading operators in each industry, teach them how to use this new tool, and then profit from the sector.
For example, in e-commerce, Silicon Minds partnered with Qianxun Holdings to establish Qianyu Intelligence, leveraging top influencers to promote digital humans. In entertainment, they signed agreements with Yuehua Entertainment.
Why have digital humans exploded in short videos/livestreaming? Sima Huapeng, founder of Silicon Minds, explains that the one-to-many interaction model of short videos/livestreaming hits the golden ratio in cost-effectiveness: "Using 200 A100 GPUs to produce content for 1,000 people is different from interacting with just one person."
This isn't the endgame for digital humans. Beyond short videos/livestreaming, AI movies and AI celebrities are even more lucrative—"creating a Wang Yibo who interacts one-on-one with fans daily."
This is the direction where digital human companies can make even more money.
Finding a clear commercialization path and achieving profitability was once considered a pipe dream in the digital human industry just a year ago. There was only burning money.
A few years ago, digital humans were more commonly referred to as virtual humans. When Meta AI sparked the metaverse craze, '3D virtual humans,' seen as the original inhabitants of the metaverse, became a highly sought-after entrepreneurial direction, with tens of thousands of startups emerging nationwide.
There are two technical routes for 3D virtual humans: one is animated digital humans, which have low rendering requirements but stiff body movements and facial expressions, with the only advantage being scalability; the other is hyper-realistic virtual humans, created through high-cost motion capture and animation to produce a more human-like and expressive virtual character, exemplified by the first-generation virtual idol Luo Tianyi.
The second type, the more human-like hyper-realistic 3D virtual humans, is the direction chosen by companies with both capability and ambition. However, its defining characteristic can be summed up in one word: 'expensive.'
Creating a 3D virtual human involves scanning, modeling, rigging, animation, and rendering. Using traditional CG-level production methods, it takes 2-3 months and costs between 1 to 2 million RMB to produce one digital human.
The more refined the 3D virtual human, the higher the cost. According to Nvidia's official blog, the 14-second video featuring Jensen Huang's virtual appearance at a launch event involved 34 3D artists and 15 software engineers collaborating, totaling nearly a thousand work hours.
However, its commercial applications are extremely limited. These 3D virtual humans often exist as IPs, directly targeting the to C market. On October 31, 2022, Liu Yexi gained 3.63 million followers overnight with her 'demon-hunting' theme, but the bad news was that operational costs also multiplied. According to previous reports from 'Jiazi Guangnian,' operating a virtual human for a year costs 8 million yuan.
Due to the high costs of 3D virtual humans, attention has shifted to 2D digital humans generated by AI, which are one dimension lower. The 2D approach produces images closer to real humans, with faster replication, lower costs, and lower operational barriers.
Silicon Minds invests hundreds of millions annually in digital human R&D, a process that has repeatedly excited founder Sima Huapeng. The data collection costs required to generate 2D digital humans have significantly decreased, with costs visibly dropping: from initially needing 30 minutes of video data, to 10 minutes, then 5 minutes, gradually shortening to one minute—or even just a single photo. 'This is a very promising curve,' he remarked.
While costs have decreased, the generated results have improved. Once, the R&D team sent a refined digital human video to Sima Huapeng, who angrily asked the team lead, 'Why are you sending me a video?' The response—that it was AI-generated—left him utterly astonished.
However, the mere reduction in costs did not help Silicon Intelligence pave a profitable path.
"How popular it is today reflects how coldly it was received before," remarked Sima Huapeng during a media interview this year, reflecting on the company's journey.
The digital human business showed little promise in its first two years of development, forcing Silicon Intelligence to sustain its team through custom projects and government contracts. In 2021, the company sold only a few thousand units, and by 2022, this number had merely grown to tens of thousands.
Silicon Intelligence did not lack effort in promotion. A team of dozens reached out daily to over 200 short video influencers via Douyin's ad engine, offering incentives: "Would you like a free digital human creation?"
Responses were either silence or rejection after learning more. "It was simply too novel," explained a member of the promotion team. At the time, Silicon Intelligence priced each digital human at just 2,000 yuan, often with a 30% discount, yet interest remained scarce.
"We were rejected hundreds of thousands of times," said Sima Huapeng. Internally, the company still holds contact details for over 100,000 Douyin KOLs—a relic of past efforts compared to the current landscape.
The essence of digital humans lies in capturing the nuances of speech—lip movements, expressions, and facial muscle details—feeding this data to AI for learning, enabling the generation of more expressions and actions to create a new "person." Digital humans represent one of the hottest applications in the current AIGC wave.
This relies on the maturity of machine vision technology, which precisely captures lip movements and facial details when a person speaks, as well as NLP (Natural Language Processing) technology, which enables digital humans to understand user intent, generate responses, and maintain coherent conversations.
Its operational principle is consistent with ChatGPT, both relying on sophisticated algorithms, powerful computing capabilities, and high-quality, extensive databases.
Before the release of ChatGPT, Silicon Minds passed the first stage of the Turing test for digital humans: it was impossible to distinguish between a real person and a digital human with the naked eye.
The improvements brought by AIGC are evident. Lin Huijie, CTO of Silicon Minds, told "Jiazi Guangnian" that a year ago, their digital humans required about a week of training by engineers after absorbing video materials. Now, with large models trained on video data from 400,000 people, AIGC can generate content in minutes, with costs decreasing exponentially.
Data accumulation has made digital humans more realistic, smoother, and more understandable. A salesperson from Silicon Minds mentioned that two years ago, clients would complain that digital humans didn’t look real, but now such feedback is almost nonexistent.
The most significant impact of AIGC is not just a technological leap but breaking the deadlock in technology adoption.
The turning point came in late October, before the release of ChatGPT, when business blogger Liu Run showcased his digital human in his annual speech "The Power of Evolution," an event internally referred to as the "Liu Run Moment" at Silicon Minds.
With the release of ChatGPT, Liu Run's endorsement spread like scattered seeds. Silicon Minds quickly raised its previously low prices, which were set due to unfamiliarity with market demand, and orders flooded in. "It exploded overnight. Once this person endorsed it, their entire community embraced the product and was even willing to promote it," said Sima Huapeng.
Platform tolerance acted as another catalyst. After the pandemic, Douyin increased its focus and investment in local lifestyle content. Industry observers noted that the platform's stance on digital humans shifted from "immediately banning accounts that livestream" to "selectively banning accounts" earlier this year.
The most suitable scenario for digital human livestreaming is the "goods-over-people" local lifestyle sector, which provides fertile ground for the growth of digital livestreaming rooms.
The number of digital humans on Douyin has grown the fastest. It wasn't until October 2022 that Silicon Minds created its first livestreaming digital human. Half a year later, Sima Huapeng proudly announced on multiple occasions, "There are 50,000 Silicon Minds digital humans livestreaming every day, generating hundreds of millions in sales revenue."
Silicon Minds provided a compelling case study: During the 2023 Spring Festival, a Hangzhou local services company, facing staff shortages, had two operators manage eight digital human-powered livestream rooms simultaneously, generating tens of millions in revenue during the holiday period.
While lacking detailed narrative, this story of two people operating eight livestreams—achieving cost reduction, efficiency gains, and holiday-season millions—became a viral success story demonstrating digital humans' wealth-creating potential in livestream commerce.
Previously uninterested parties—Douyin KOLs, e-commerce merchants, and even Qianxun Chairman Dong Haifeng—began reaching out. Business inquiries flooded company phone lines, even reaching administrative departments through the official website.
Sales representatives noticed a dramatic shift. Where they once had to explain digital humans' definition and benefits, clients now immediately ask about pricing, customization, and delivery timelines.
The company name itself became a talking point. Previously, promoters faced questions like "Is Silicon Minds about silicone rubber?"—referring to the silicon element in chips. Overnight, clients praised the foresight in naming after this once-obscure concept.
In Hangzhou, Zhan Jian first noticed the digital human hype in March during their transition period. A late-night viral video—featuring 70 screens displaying digital human livestreams in an empty office set to "The Ring" soundtrack—garnered 600,000 views with the caption "No humans left, only digital replacements."
Since then, Zhan hosts three visitor groups daily at their digital human display wall, often juggling two phones simultaneously during peak periods.
Three months later, Zhan Jian sensed a whiff of competition. Agents in Yiwu began asking him, 'What advantages do your digital humans have?'
More people started comparing prices after inquiries. Zhan Jian's price was 2,980 yuan to clone a digital human, while the market price had already dropped to 300 yuan per unit.
The anxiety spread by AI has turned digital humans into a money-making tool for some.
Before hosting an AIGC conference, Xu Zhibin, founder of Jianshi Technology, distributed 325 questionnaires about AI. Among them, 90% of respondents expressed willingness to pay to learn AI technology, with founders particularly anxious about AI disrupting their companies' advantages.
This mindset has been exploited as a sales pain point by digital human sellers. A practitioner in the digital human industry promoted digital human courses in a live broadcast. When businesses asked how to address the post-pandemic issue of insufficient offline foot traffic, the uniform answer was to use a new weapon—'AI digital humans'—to provide novelty. This was followed by a sales pitch: 'For less than 10,000 yuan, you can experience and use AI, which is negligible for bosses. If the boss doesn’t use AI, how can the employees?'
Digital humans are indeed a valuable tool in certain scenarios.
In 2022, Zhan Jian assisted a socket manufacturer attempting to cut livestreaming costs. When store broadcasts were just emerging, the company hired two complete livestreaming teams (including hosts, control operators, and managers) at a cost of over a million yuan to build a livestreaming matrix. They later decided to experiment with digital humans replacing some real hosts. After three months of trial broadcasts, Zhan found that while digital human broadcasts generated slightly lower metrics than human hosts, the cost difference was negligible: digital humans cost less than 1/10th of human hosts.
This socket manufacturer subsequently adopted this approach extensively, operating dozens of digital human livestream accounts through Zhan's service. These accounts broadcast for extended periods daily, up to 40 hours, ensuring steady revenue as long as transactions occurred.
Another case involved a cultural tourism site that initially hired human hosts for livestreams, generating 100 million views but zero ticket sales. After allocating part of their budget to digital humans, they now see consistent daily revenue.
However, digital humans' value in livestreaming primarily manifests in store broadcasts and existing IP channels. They can maintain revenue models and reduce costs by replacing human hosts, but cannot create profitable livestream channels from scratch.
In practical implementation, the myth of digital humans being an easy money-maker has been debunked. Sima Huapeng compares it to "thinking buying machines alone can generate profits - this unmet expectation is the industry's biggest pain point." He believes effective AI in livestreaming requires six essential elements: technology, operations, supply chain, IP, finance, and accounts - none can be omitted.
A digital human company founder warned agents during a sharing session: The real need is using digital humans to reduce costs and increase efficiency; the false need is expecting digital humans to salvage situations where even human hosts failed.
In short videos, AIGC digital humans can almost pass as real people by clearing the Turing test, helping many hosts avoid repetitive recordings. However, in more interactive live streaming scenarios, these digital humans still lack the ability to convincingly engage. Factors like asking questions, listening to the host's voice, or observing significant movements can reveal whether the live streamer is a digital human or not.
Digital humans generate revenue by extending broadcast hours. While a human host might stream for 6 hours, a digital counterpart can run for 8-12 hours, with some even lasting over 40 hours. However, prolonged digital broadcasts require careful handling of platform review systems. Low-quality digital humans won't be promoted by platforms, resulting in viewership as low as one person per second.
Lin Huijie, CTO of Silicon Minds, states that achieving an ideal live streaming environment depends on digital humans passing the Turing test in live settings. This requires rich expressiveness and coordinated body movements, demanding significant computational power.
Beyond internal operations, external factors complicate matters—platforms are cracking down on low-quality content. On May 17th, Douyin updated its policies: digital human live streams must now include real human participation. "This is like requiring a driver to stay behind the wheel in autonomous vehicles," explains Sima Huapeng.
The industry has developed countermeasures. For instance, Qianyu Intelligence employs an "Avatar" model where a real person provides real-time scripts for the digital human, with AIGC generating lip-sync and expressions. Purely AI-generated scripts risk being flagged as pre-recorded content, leading to account bans.
Zhan Jian shares strategies to evade platform scrutiny, such as intermittent streaming during the high-risk 7-15 day account setup period—"stream for 2 hours, go offline, then restart"—or creating 10 clips from one live session and airing them in different orders across days.
Even so, digital human accounts being suspended is a common occurrence. During customer testing, the temporary suspension rate (1-3 days, no permanent suspension) for 100 accounts was around 30%. Among the 30 accounts that could be promoted by the platform to generate daily transactions, only 15 succeeded.
The distance from AI application to live streaming is similar to how machine guns were invented but remained unused for decades before being widely deployed on the battlefield. There is often a significant gap between the invention of a new technology and its widespread adoption.
"From a profit perspective, it's a good business, but from the standpoint of technology for good, it's not," said Sima Huapeng. Creating AI is not about destroying humanity, but some industry malpractices are doing just that.
Melanie Mitchell mentioned in AI 3.0 the striking words of AI researcher Pedro Domingos: "People worry that computers will get too smart and take over the world, but the real problem is that computers are too stupid and they’ve already taken over the world."
Sima Huapeng shared their experience with telephone robots, which had a similarly lucrative business model as today's digital humans. Countless companies flooded into this sector until the government intervened to regulate spam calls, and telecom operators stopped approving lines for robots. The result was, "Everyone killed the industry."
So, what about digital humans?