China’s AI idol craze: Fad or future?

(Edited and refined by Josephine Hong, with the assistance of AI translation.)

In June 2026, The Laid-off Girl (《被裁掉的女孩》), a short drama generated entirely by AI, premiered in mainland China. Within a month, it had attracted more than 200 million views, becoming a breakout hit of the summer and drawing industry attention.

The story follows Fang Taozi, a small-town girl who moves to the fictional city of Shangnan to pursue her dream of becoming a model. After repeatedly being sidelined at work for refusing to submit to unspoken rules, she ultimately grows and transforms through her own efforts. More unexpectedly, the virtual heroine Fang became a hit. 

Able to cry, laugh, film vlogs and even play up romantic pairings, Fang has established a convincing sense of being a real person through her vivid personality. Together with lifelike performances and settings, the AI production is easy to forget once viewers are immersed in the story, seeming almost indistinguishable from a drama filmed with real actors.

The sense of realism has been sustained by the production team taking Fang beyond the drama itself. They set up dedicated social media accounts for her, posting snippets of her daily life and engaging with netizens. Although the short drama has ended, Fang’s “life” has not. She continues to interact with fans through social media and share behind-the-scenes stories. In just 30 days, “her” personal account gained 470,000 followers and received more than 31 million likes.

Scandal-free celebrities

The earning power of this newly minted AI influencer should not be underestimated either. Top Douyin influencers with 10 million followers typically charge between 100,000 RMB and 200,000 RMB (about US$14,800 to US$29,700) for a 60-second advertisement. Fang Taozi’s rate is quoted at 258,000 RMB, and brands have actually paid up. On 31 July, Fang released her first commercial collaboration video, featuring a product placement for coloured contact lenses. It marked the formal entry of AI-generated virtual characters into commercial brand advertising.

For brands, the most immediate appeal of AI virtual humans is that they eliminate many of the uncontrollable variables associated with real people. Real celebrities can become embroiled in controversy at any time because of inappropriate remarks, scandals in their private lives or the collapse of their public image. Any negative incident can render a brand’s marketing budget worthless. 

Meanwhile, virtual humans do not need brands to accommodate their schedules, nor do they require repeated coordination over reshoots, script changes or image retouching. Once a character has been modelled, it can remain “on duty” around the clock, with the setting, dialogue and style swiftly adapted to a brand’s needs and extended across short videos, livestreams, advertisements and other content formats. Some netizens have joked that AI entertainers will never act like divas, and it is just a matter of time before they replace real actors.

Yet being controllable does not mean being risk-free; the problems have merely taken a different form. As AI becomes increasingly capable of replicating real human characteristics, issues surrounding portrait rights, copyright and intellectual property (IP) ownership will become more complex. Fang Taozi previously sparked discussion for resembling Japanese actress Nana Komatsu. Meanwhile, the commercial value of virtual humans has yet to be fully proven. Whether audiences will buy merchandise, show support and even form lasting emotional bonds with them in the same way they do with real idols remains for the market to determine.

From digital puppets to AI with brains

Virtual humans are not new. From virtual singer Hatsune Miku and Instagram virtual influencer Lil Miquela to Japanese virtual model Imma, people have long been drawn to using digital tools to create characters that do not exist in the real world. Singapore, too, has had virtual influencer Rae, who presents herself as a 25-year-old Singaporean on social media, drinks kopi peng (colloquial term for iced coffee), visits local neighbourhoods and collaborates with real brands.

But early virtual humans had an obvious limitation: they looked human, but could not “think” for themselves. Every image was the product of modellers, animators and image editors, while every social media caption was written by human operators. Virtual characters were more like digital puppets controlled by people behind the scenes.

Generative AI has changed this. Large language models give virtual humans the ability to converse, voice models provide them with voices, and generative video creates their facial expressions and movements. Combined with knowledge bases, long-term memory and agentic AI capabilities, digital humans are evolving from a “talking face” into AI interfaces that can understand and answer questions in real time and even carry out tasks.

In March 2026, AI avatar company D-ID launched a new generation of ultra-high-fidelity digital humans, V4 Expressive Visual Agents. Through real-time links with large language models, it has a conversational response delay of under 0.5 seconds and can alter its voice, facial expressions and lip movements according to the context. Video characters that previously could only play pre-recorded content can continue interacting with viewers in real time after the video ends, with knowledge, memory and agentic AI capabilities. Put simply, people used to watch virtual humans; now they are beginning to interact with them in earnest.

A digital double for busy people

Similar applications are already entering the corporate world. In 2026, video technology company Kaltura unveiled an eSelf digital double of its CEO, Ron Yekutiel. The virtual CEO, equipped with Yekutiel’s voice, appearance and communication style, handles some staff communications, company updates and client engagement. There is only one real person, but a digital double can enable someone to appear in different places simultaneously, like the Monkey King (Sun Wukong) turning a single hair into 70 copies of himself, and speak different languages to different audiences.

For celebrities, corporate leaders and content creators with packed schedules, this opens up a range of possibilities. At another level, human identity itself appears to be becoming a digital asset that can be replicated, expanded and licensed for use.

Virtual humans already put to work

If Fang Taozi has given the Chinese public a first glimpse of an “AI star”, Singapore’s current focus is more on turning virtual humans into AI employees. Pantheon Lab, an AI company listed with Singapore’s Infocomm Media Development Authority (IMDA), has developed an AI-agent infrastructure for virtual humans that connects generative AI and virtual-human interfaces to corporate back-end systems.

This enables digital humans not only to answer questions but also to help users complete tasks. These digital ambassadors can be deployed in everyday spaces such as post offices and museums, or serve as virtual customer service officers at bus interchanges and MRT stations. When people make enquiries, they no longer see a row of text on the screen in response; instead, they are greeted by a virtual staff member with a face, voice and AI brain.

Another company listed in IMDA’s innovative technology company directory, Topview, is bringing AI virtual humans into the content industry. It can generate promotional videos with a realistic human feel and rapidly create versions in different languages, without the need to find actors, film crews or influencers. Its content includes product advertisements, social media videos, short films and animation. Since its establishment in 2022, the company has served more than 2,000 users in over 120 countries.

The broader industry trend is in step with Singapore’s current push for agentic AI. Since 2025, the government has worked with Google to establish the AI Agents Sandbox to test agentic AI and explore its applications in the public sector. In January 2026, Singapore launched the world’s first government-led regulatory framework for agentic AI, helping companies and institutions take measures to reduce deployment risks. It emphasises and ensures that, even as intelligent agents gain increasing autonomy, humans must ultimately remain responsible for how AI operates.

‘Realistic’ does not mean real

The virtual human ecosystem is a new blue ocean, rich in both opportunities and challenges. The real challenge posed by AI virtual humans is not whether they resemble real people, but how society should redefine “authenticity” and “responsibility” when people begin to trust, consume and rely on them despite knowing they are not real. If Fang Taozi were to recommend a skincare product, could she say that she had used it and found it effective, when she has no real skin and cannot have direct sensory experience? The commercial value of virtual humans lies in their ability to imitate the sense of closeness offered by real-life influencers, but such closeness is not the same as genuine experience.

As technology advances, regulations are also beginning to set new boundaries for AI virtual humans. Draft rules on the management of digital virtual human information services introduced in China in April 2026 require clear labelling of virtual human content. They also prohibit the use of others’ personal information to create digital humans without consent, and specifically restrict virtual humans from forming virtual intimate relationships with minors.

Fang Taozi’s popularity suggests that a shift in emotional connection has already begun. Netizens know she does not exist, yet they still look forward to her updates and discuss her storylines and romantic relationships. Knowing that she is AI has not stopped people from investing their attention in her.

After years of technological development, computer-generated people are becoming ever more lifelike. Beyond whether people can tell the real from the fake lies another question: even when they know the other party is not real, are they still willing to form a relationship with it? 

Fang Taozi may be only the beginning of the virtual era. The next encounter with an AI virtual human may come while asking for directions at an MRT station, listening to a museum guide, undergoing corporate training or dealing with a company’s customer service manager. The face on the screen smiles, listens and answers questions, but there may not be a real person behind it. As these “non-human” characters begin to play a genuine role in human life, the next question is no longer how closely AI can resemble humans, but whether people are ready to live alongside them.

This article was first published in Lianhe Zaobao as “能哭会笑还接广告 AI虚拟人开启吸金时代”.

Related: Chatting alone? What AI companions reveal about human relationships | AI micro-dramas do not understand people. Are we starting not to either?