红字

你没猜错，微信又在灰度测试新的 AI 功能了。         Still frame from an AI-generated short video depicting the Chinese-style "heavenly court" created by creators Qiguang and Xueli Xianzong (screen names) Photo: Courtesy of QiguangWhat if one person and a computer could create something that wows the world?
Meet Qiguang, a 1991-born eyeglasses store owner whose AI-generated Chinese celestial realm racked up 5 million views overseas.
He is part of a rising movement in China: the "handcrafted economy," in which low-cost AI technologies and creativity are opening a new path to entrepreneurship.A white jade corridor stretches deep into a sea of clouds. The towering celestial palace rises above tiered upturned eaves, and red-crowned cranes glide between floating sanctuaries. 
Shared online by viewers without subtitles or dialogue, this AI-generated short video racked up more than 5 million views across overseas social platforms in less than four days, attracting audiences worldwide.
Many overseas viewers assumed the atmospheric visual sequence came from a professional film studio or seasoned 3D animation team. 
However, its creator, known online as Qiguang, leads a far more ordinary life. Born in Chengdu, Southwest China's Sichuan Province, in 1991, he now runs a physical optical shop. In his spare time between serving the customers, he taught himself AI image and video generation, painstakingly crafting the viral vision of a Chinese celestial realm.
Qiguang's overnight international fame is no isolated internet success story. It signals the spread of a new trend dubbed the "shoucuo" economy or handcraft economy. The term describes the process of bringing ideas to fruition by hand on a small scale with simple tools, relying mainly on individual skill or manual labor rather than dedicated industrial production lines.
On Chinese social media, the hashtag "Hand-Crafting Everything" has surpassed 100 million views. In app stores, solo developers are launching AI-powered tools that compete directly with products backed by tech giants. This movement, defined by low costs, high creative freedom and flexible, fragmented workflows, is reshaping the national imagination of entrepreneurship.
Two people, one vision of heaven
Few viewers of Qiguang's viral celestial video realized that behind its ethereal beauty stood not a studio of dozens, but just two people, and hundreds of discarded AI attempts.
Qiguang told the Global Times that he had long been fascinated by the grand, ethereal landscapes of Chinese mythology - the celestial palaces and floating realms described in ancient texts. Then he discovered the work of another creator, Xueli Xianzong (screen name), whose AI-generated visuals captured a distinct Eastern aesthetics and minimalist blank-space composition.
"I realized AI could break down the barriers for ordinary people," Qiguang said. "The images I'd held in my mind for so long could finally become visible."
Xueli Xianzong had arrived at a similar realization earlier. For years, he had wanted to visualize the celestial palaces of Chinese fantasy fiction, but traditional 3D modeling and live-action filming were prohibitively expensive for an individual. AI generation tools changed that. He began spending his free time learning text-to-image and image-to-video technologies, carving out a niche in the relatively obscure genre of Chinese fantasy visuals.
The two eventually became creative partners, though their collaboration is unusually loose.
"Mostly we create independently, then share our finished work for critique and inspiration. We absorb aesthetic ideas from each other's visuals and incorporate them into our own," Xueli Xianzong said. "This is how most creators in this space collaborate right now."
But if their creative partnership was remarkably uncomplicated, their struggle with AI itself was anything but straightforward. The process behind a video like "Chinese Heavenly Court" is far more painstaking than audiences have imagined. Qiguang broke it down: First, conceptualize the overall atmosphere and frame each shot's camera movement. Second, use an AI image generation tool to generate high-quality static images for each frame, endlessly tweaking prompts to control architectural details, color palettes, lighting and composition. Third, feed the selected images into AI video-generation models to add dynamic lens effects, like horizontal camera panning, drifting clouds, flowing sleeves. Fourth, sort through the output, discard distorted or broken footage, edit the clips together and finally choose music to match the mood.
"AI frequently produces garbled architecture, deformed figures or unnatural motion," Qiguang said. "You envision something perfect in your mind, but after dozens of generations, you still haven't transformed your vision into reality." 
After two years of self-learning, Qiguang regards his biggest breakthrough as escaping generic mass-produced ancient Chinese aesthetics. To him, the handcraft economy differs fundamentally from traditional entrepreneurship at its starting point. 
"You can begin with just one person and a single computer, turning ideas directly into content or products. Heavy upfront investment disappears, yet every tedious task, every round of trial and error falls solely on the creator," Qiguang said.
"The AI-assisted handcraft model removes many steep technical barriers. If you possess taste and imagination, you can experiment in your spare time," he noted with a beam of proud.
Guo Yi, a professor at the School of Journalism, Chongqing University, told the Global Times the handcraft economy can be more accurately described as "intelligent creation economy" or "geek creation economy": Human intellect directs computing power, and creativity guides automated tools. In many ways, the movement mirrors the global maker economy. 
"Yet generative AI brings transformative changes beyond merely lowering entry costs for creative industries. It is restructuring cost frameworks and production relationships across creative sectors," Guo said.
A fast-growing community  
    A visitor inquires about AI-powered entrepreneurship solutions at an industry fair hosted by local authorities in Guangzhou, South China's Guangdong Province on March 15, 2026. Photo: VCGIf Qiguang and Xueli Xianzong build dreamlike visuals in the creative clouds, 34-year-old Xu Ting "handcrafts" practical digital products grounded in real demand. 
Xu still vividly remembers, in January 2025, publishing a simple post on the social media platform Xiaohongshu: "I want to build an AI app that chats with elderly users and helps them compile life memoirs. Anyone want to join?"
Under conventional internet industry logic, the idea sounded unrealistic. Developing an app once required large teams and millions of capital. But AI rewrote these rules. 
Xu previously worked in the energy sector with no coding background. Using domestic large language models including DeepSeek, she assembled a working product prototype block by block. Without dedicated product managers or UI designers, she single-handedly completed market research and product design.
"In the past, many good creative concepts were abandoned simply because their costs were too high. Today, AI has driven down model token costs, making experimentation accessible to ordinary people and drawing growing numbers of newcomers into this space," Xu told the Global Times.
The day after her recruitment post went live, a flood of private messages confirmed genuine market demand.
Li Ning, a professor at Tsinghua University's School of Economics and Management, pointed out that traditional entrepreneurship demands heavy investment in research, staffing, supply chains and market testing, discouraging many innovators. 
"AI compresses these costs. Small teams working through independent handcrafted development can achieve outputs that once required hundreds of workers. This has also given more people the courage to start their own businesses." Li said.
Beyond gaps in costs, "handcrafted" teams follow different organizational and collaborative frameworks from established enterprises. Li told the Global Times that inside major companies, projects are pushed to deliver steady, foreseeable returns from conception through rollout. This type of environment creates barriers for disruptive ideas to flourish. Handcrafted teams, however, benefit from their lean size: They can shift direction quickly and iterate through constant trial and error. 
"This has also given more people the courage to start their own businesses," he said.
Xueli Xianzong identified four core distinctions between the handcrafted economy and traditional entrepreneurship: low barriers to entry; creativity as the core while tools remain carriers; flexible collaboration without formal incorporation or fixed teams; and low trial-and-error costs, allowing creators to explore niche verticals without financial ruin.
According to Alibaba Cloud, the cloud computing arm of China's internet giant Alibaba, as of November 2025, nearly 23,000 AI applications have been built within the innovation zone on its ModelScope open-source platform, and roughly 95 percent of them were developed by individual creators.
Data from the "2026 China One Person Company (OPC) Industry White Paper" and sector reports show more than 16 million registered OPCs across China, as of April 2026.
Navigating headwinds for the creators
Yet if AI has flattened the technical playing field, what truly determines success? For those riding the wave of the handcrafted economy, the AI era offers a double-edged promise: unprecedented freedom to turn ideas into reality, paired with formidable obstacles that test resilience at every turn.
For Qiguang, it is the ability to perceive beauty: innate aesthetic judgment and imagination. 
Xu argues that the most vital asset for new entrants is not capital, but sharp insight into unmet real-world needs. "AI excels at execution. Figuring out what to build is far harder than building it," Xu said. 
The true competitive edge lies in translating technical capability into tangible user value. "Identify genuine underserved demands and turn technological capacity into value for users. Technology determines whether you can build a product. Understanding users, patience for iterative refinement and long-term trust determine how far a micro-venture can travel," she noted. 
But even with the right instincts, the path from idea to market is anything but smooth. The explosive growth of handcrafted products brings steep challenges on that journey. 
Xueli Xianzong noted AI enables cheap copying and imitation, while individual creators face exorbitant legal costs to defend their works. When infringement costs remain far lower than enforcement costs, creative enthusiasm inevitably fades.
Beyond intellectual property, creators must escape overreliance on fleeting online traffic. Qiguang observed platform algorithms tend to prioritize snappy, conflict-driven short content, while slow, culturally rich expression struggles to gain traction. He refuses to churn out formulaic commercial drama series chasing view counts, choosing instead to deepen grand storytelling. 
"The handcrafted economy is no shortcut to quick wealth. It demands patience through periods of little reward," Qiguang said, noting that new participants need sufficient time and a willingness to endure repeated trial and error.
From an app development perspective, Xu highlighted another tension: balancing customized service and standardized workflows. If every project relies entirely on manual handcrafted labor, founders risk being trapped in high-intensity, low-profit self-employment. The sector needs mature workflows, delivery benchmarks and pricing frameworks. Tasks suitable for standardization can shift to AI, while human empathy and critical judgment stay central. Privacy and ethical safeguards also grow critical, especially for services handling personal memories and sensitive information.
Guo views inconsistent quality standards as typical growing pains, which market filtering mechanisms will gradually resolve. Delayed intellectual property frameworks and emerging governance challenges, however, require thoughtful ongoing policy attention, he said.
Authorities across China have begun rolling out supportive measures to address these obstacles. Over 20 cities have launched targeted policies for OPCs. The Beijing Economic-Technological Development Area offers streamlined business registration services for solo entrepreneurs, delivering business licenses in as little as 30 minutes. Qingdao in East China's Shandong Province rolled out a specialized platform for solo entrepreneurs, delivering end-to-end support for AI innovators from technical development to revenue generation.  
At the same time, a network of OPC communities has taken shape. As of May 2026, 426 such communities operate across 28 provincial-level regions and 65 cities. These hubs provide physical workspace alongside networking events, resource matching and mentorship, connecting previously isolated solo creators into a connected ecosystem.
    Still frame from an AI-generated short video depicting the Chinese-style "heavenly court" created by creators Qiguang and Xueli Xianzong (screen names) Photo: Courtesy of QiguangNow, Qiguang continues running his optical shop full-time, pursuing AI creation only as a side project with no immediate path to full-time work or large-scale earnings. Both Qiguan and Xueli Xianzong say they will consider full-time creative work only when industry conditions and monetization models mature further. Still, they view handcrafted creation merely as a starting point rather than a final destination, and remain optimistic that this grass-roots creative movement holds vast untapped potential ahead.
Huo Minhan contributed to this story
。这次灰度测试的功能叫做「AI 修图」，它会出现在微信聊天的图片窗口中，和原本的「查看原图」按钮并列把底下一排塞满：无论是你发的图还是接收到的图，在预览窗口点击「AI 处理」之后就可以看到 3 大类共 11 个小类的预设功能：常见功能比如美颜、加滤镜、翻译文字、自动打码和超分等等一应俱全，点击就会跳转到微信小微对话框中开始工作：至于小微修图的质量嘛……只能说还行。经过测试，小微在美肤、扩图、调整比例和消除杂物之类的功能上表现尚可，虽然能看出破绽，但对需要边聊边做的轻量编辑完全足够了——但如果你也被灰测到了微信主界面左上角的小微入口的话，那直接进入小微对话框把要求发给它，和使用那些快捷开关其实是完全一样的：不过目前看下来，小微的修图能力在「编辑」和「信息提取」方面还算不错。可一旦涉及重绘或者生成新内容，就比较吃用户的提示词水平了——通过直接询问小微，我们得知它目前使用的还是微信定制的 WeLM 模型，大概率是之前已经在对话、小程序里面上线的 WeLM-80B 版本。而纵观 2026 年至今，我们可以清晰地看到一件事：哪怕像微信这种「牵一发而动全身」的超巨型 App，也开始非常积极地拥抱 AI 功能了。从 3 月首次出现微信研发 AI Agent 的消息、到 6 月初确定使用形式，再到小微的灰测和拓展，我们可以梳理出微信的 AI 时间轴：重点在于，微信没有急着再造一个豆包或者 ChatGPT，而是在把尝试把一个综合模型能力和入口拆开，塞进微信原本已经存在的功能里。目前可见的就是，除了主页左上角的小微聊天框之外，「小微功能」其实已经渗透进了微信的方方面面——而 WeLM 底层模型也是围绕这种「方方面面」来设计的。比如腾讯在最新的财报中首次披露：微信团队定制的 WeLM 在设计时就重点考虑了微信场景和推理效率，目前公开的 WeLM-HD4-80B 和 WeLM-HD4-617B 总参数量分别达到 800 亿和 6170 亿，但实际激活的参数只有 30 亿和 230 亿。

图｜X @Weixin_WeChat微信团队还提出了诸如 Hidden Decoding 之类的技术，通过将每个 Token 扩展成多个内部计算流，提高单 Token 的计算量，同时用 Stream-Factorized Attention 控制额外的注意力开销。这些技术参数背后，其实不是腾讯的技术边缘，而是微信最现实的问题：算力成本。截至今年 6 月，微信及 WeChat 合并月活已经达到 14.39 亿。小微一旦全面进入搜索、图片、朋友圈、公众号乃至小程序场景……每一次推理成本乘上十亿级用户规模，都会变成一个天文数字。

图｜财联社所以 WeLM 如此强调推理效率，表面能力看上去往往只是「差不多得了」，其实不是模型限制，而是微信的巨大运营成本压力。更重要的是，相比腾讯的 WorkBuddy，微信小微走出了一条完全不同的推广路径。

二 | 微信手握 14 亿月活，以及无数人每日高频使用的功能场景，根本不需要去「给模型造 App 入口」，反而是让小微尽可能淡化唯一入口的概念、拆分进具体功能里。然而世界是物质的，哪怕是微信这种地位的 App，也无法完全避免谈钱的问题。根据财报相关资料，腾讯最新季度资本开支达到 528 亿元，同比增长 176%；自由现金流则转为负 138 亿元，其中重要原因就是 AI 相关的算力采购和预付款。腾讯披露，这些基础设施正在支持模型、WorkBuddy、CodeBuddy，以及微信 AI 项目的训练和推理。

图｜彭博社另一个数字更直接——腾讯当季 Non-IFRS 营业利润为 756 亿元，如果剔除元宝、WorkBuddy、CodeBuddy、小微等 AI 产品的影响，实际上的利润则是 861 亿元。也就是说，按照腾讯自己的财务口径，新 AI 产品目前正拖累着腾讯集团约 105 亿元的季度经营利润。这也是腾讯接下来必须回答的问题：AI 已经有产品、有用户、有使用量，但什么时候真正变成一门生意？

图｜新浪财经微信上线 AI 修图这件事本身并不大。真正的重点是，两个月前还藏在主界面角落里的「小微」，正在逐渐成为一个 Gemini 一样「无处不在」的 AI。继续沿着这条路线走下去，小微不会成为一个「AI 小程序」，而是微信的 AI 能力本身。