Researchers studying AI image generators like Midjourney, DALL-E, and Gemini have discovered that these tools may be reinforcing narrow beauty standards in ways that mirror China's controversial 'Internet celebrity face' aesthetic. A composite analysis of AI-generated images of young Chinese women revealed consistent facial feature patterns across all three platforms, raising questions about whether generative AI is truly democratizing beauty or simply encoding existing biases into algorithms .

What Is 'Internet Celebrity Face' and Why Does It Matter?

China's digital economy has long been shaped by "wanghong," a term meaning "net red" or "hot online," referring to influencers who dominate livestreaming platforms and social media. At the core of their influence is a highly standardized look known as "wanghonglian," or "Internet celebrity face." This aesthetic is defined by specific features: a pointed chin, European-style double eyelids, and a high-bridged nose .

The prevalence of this look is staggering. Among the 100 top-rated female livestreamers on China's money-spinning platforms, 93%, 96%, and 87% respectively possess the "Internet celebrity face" aesthetic . This standardization has become so entrenched that it sparked intense debate in 2016, when it became "the most discussed buzzword of the year" in China, partly because the trend reflects broader concerns about a lookist society and mounting social injustice .

Are AI Image Generators Perpetuating the Same Beauty Standards?

The research examining Midjourney, DALL-E, and Gemini suggests a troubling parallel. When researchers conducted a zero-shot analysis, meaning they asked the AI tools to generate images of young Chinese women without providing specific facial feature instructions, the tools produced remarkably similar facial characteristics across all three platforms . This finding raises a critical question: are generative AI systems learning and reproducing the same standardized aesthetics that have dominated China's digital culture?

The implications extend beyond aesthetics. If AI image generators are encoding these narrow beauty standards into their outputs, they could reinforce existing biases and limit the diversity of representations available to users worldwide. This is particularly concerning given that these tools are increasingly used in commercial applications, from marketing to content creation.

How AI Is Transforming China's Beauty Industry

• Cosmetic Surgery Integration: Smartphone-based AI tools now evaluate patients' suitability for cosmetic surgery, while large language models facilitate dialogue during consultation processes, personalizing the surgical planning experience.
• Digital Retail Applications: AI technology is permeating brick-and-mortar beauty retail sites in Chinese cities through digital touchpoints that offer smart diagnostics and skin analysis powered by big data, with major retailers like Sephora testing AI-driven product recommendations.
• Facial Modification Software: Apps like Meitu have evolved from simple rule-based photo filters to machine learning-powered tools that can slim faces, narrow jaws, and enlarge eyes, making digital facial modification accessible to millions of users.

The shift toward AI-driven personalization in beauty represents a significant departure from the one-size-fits-all approach of the past. Wu Xinhong, CEO of Meitu, China's premier beauty filter app, acknowledged this transformation in a 2021 interview .

"In the past the company's products had unified everyone's aesthetics, but nowadays more and more flowers are blooming, and aesthetics are becoming more open and inclusive. We're gradually making everyone understand that their own beauty is unique," stated Wu Xinhong, CEO of Meitu.

Wu Xinhong, CEO of Meitu

This statement reflects an industry-wide pivot toward customization and individuality. However, the research on generative AI tools suggests that the promise of personalization may be complicated by algorithmic bias. If Midjourney and similar tools are generating images with consistent facial features regardless of user input, the technology may be creating a new form of standardization rather than enabling true diversity.

The Paradox of AI-Driven Personalization

China's beauty industry presents a fascinating paradox. On one hand, AI and algorithmic governance are being marketed as tools for personalization and individuality, moving away from the homogenized "Internet celebrity face" aesthetic that dominated the late 2010s. On the other hand, the underlying AI models powering these tools may be reproducing the same biases and narrow standards they claim to transcend .

The evolution of Meitu illustrates this tension. When the company launched Meitu Xiuxiu in 2008, it used simple rule-based image-processing algorithms to apply filters and whiten skin with a single tap. This democratized access to Hollywood-style glamour but also standardized beauty ideals. As the company evolved to use machine learning, it gained the ability to create more nuanced and varied edits. Yet the underlying question remains: are these tools truly enabling individuality, or are they encoding new forms of conformity?

The research examining Midjourney, DALL-E, and Gemini suggests that generative AI tools may need significant refinement to avoid perpetuating existing beauty standards. As these tools become increasingly integrated into commercial applications, from cosmetic surgery planning to influencer marketing, understanding their biases becomes crucial for ensuring that AI-driven personalization actually delivers on its promise of diversity and inclusion .