Can AI Accurately Express “Chinese Style”?

A Visual Try-On Experiment Across Cultures

As I began exploring the intersection of AI and Chinese aesthetics at Central Saint Martins, I was particularly curious about one key question:

Can AI-generated images authentically express the essence of Chinese fashion—and will global users recognize it as such?

To investigate this, I designed a small-scale visual intervention combining AI try-on technology with quick interviews. The results were surprising, insightful, and at times, culturally revealing.

The Experiment

Using Lavie AI, a virtual outfit simulation platform, I generated six images of models with different skin tones and body types wearing traditional clothing:

  • 4 Chinese-style looks (e.g. stand-up collars, frog buttons, Tang/Song silhouettes, red and white color tones)
  • 1 Japanese look (loose-fit linen kimono style)
  • 1 Korean look (hanbok-style garment)

Then, I invited five people from different national and cultural backgrounds—including classmates and family members from Colombia, Indonesia, Iran, and the UK—to review the images. Each of them was asked:

  1. Which image looks most “Chinese-style” to you? Why?
  2. Which image is the most visually appealing to you? Why?
  3. Would you be interested in clicking or learning more about the outfit or brand if you saw it online?

Interview Snapshots: Real Reactions

I collected their answers via WhatsApp chats and voice notes. Here are some excerpts from the interviews:

Simon (Colombia):
Picked image 4 as the most “Chinese” due to the setting and model appearance. Found image 5 the most visually striking because of the cultural contrast (Black model in Asian-style outfit).

Sasha (Indonesia):
Recognized image 4 for its traditional elements. Said image 1 looked “authentic and wearable

Meaghan (UK):
Noted that the frog buttons and stand-up collar on image 1 reminded her of a qipao. Liked the elegance of image 4.

Goldaneh (Iran):
Picked image 4 for its patterns and composition. Felt it conveyed strong traditional Chinese visual language.

Simon’s Mom (Colombia):
Believed image 1 looked traditionally Chinese. Appreciated its simplicity and elegance.

Key Insights

Cultural Bias & Misidentification

  • One participant mistook the Japanese outfit for Chinese, indicating how Asian styles are often visually conflated in the global imagination.
  • Another successfully identified the Korean hanbok.
  • Many respondents based their judgments not only on clothing details, but also on model ethnicity and background setting, showing how race and context affect perception.

Visual Preferences

  • Images combining traditional elements with modern simplicity were more appealing.
  • Participants appreciated subtle elegance and familiar cultural patterns (e.g., buttons, collars, color tone).

What This Means for AI, Design & Culture

AI Has Cultural Expressive Potential
Despite being algorithmically generated, LavieAI’s visuals were widely recognized and appreciated for their cultural tone—supporting the idea that AI can serve as a translator of aesthetics.

Cultural Misunderstandings Remain Common
The confusion between Japanese and Chinese outfits highlights the need for visual literacy and educational framing, especially when working across cultures in fashion.

A New Opportunity for Design Communication
This experiment also reinforced my broader goal: to help Chinese aesthetics break out of outdated or kitschy tropes, and reintroduce its elegance through modern, intelligent tools—like AI.

Next Steps

Going forward, I aim to:

  • Explore building a localized aesthetic database for more accurate cultural modeling
  • Work with LavieAI on expanding visual outputs that better distinguish cultural codes
  • Combine AI-generated images with narratives, color theory, and symbolic tags so global users can understand the “why” behind each look

Related Research & References

Would you have made the same guesses?
What does “Chinese style” mean to you in 2025?
Share your thoughts below.


Comments

Leave a Reply

Your email address will not be published. Required fields are marked *