Meta’s image generator appears to struggle with accurately depicting intermarriage.
In a report by Mia Sato for The Verge, her efforts to create an image of a mixed-race couple—specifically an East Asian man with a Caucasian woman—were met with repeated failures. Despite trying “dozens of times” using prompts like “Asian man and white wife” and “Asian man and Caucasian woman on wedding day,” the AI consistently generated images of two Asian individuals. Only once did it produce a somewhat accurate image, which oddly featured a noticeably older man alongside a young, light-skinned Asian woman.
This issue contrasts sharply with a recent controversy surrounding Google Gemini, where, shortly after its launch, the chatbot was criticized for delivering historically inaccurate depictions, such as female Catholic popes and Asian Nazi soldiers. Following widespread backlash, Google suspended Gemini's person-generating features and issued an apology.
Sato described her results as “egregious,” highlighting how generative AI can stifle creativity, saying it "imprisons" it within societal biases.
Despite attempts to communicate with Meta for comment, there was no response.
Stereotypes and Skin Tone Bias
Sato's exploration revealed that minor adjustments to prompts did little to improve outcomes—using "white" instead of "Caucasian" yielded the same results. The generator also hesitated to portray platonic relationships. For prompts like “Asian man with Caucasian friend” or “Asian woman with Black friend,” it repeatedly displayed images of Asian pairs instead.
Furthermore, the bot reinforced biases surrounding skin tone. The prompt “Asian woman” yielded predominantly light-complexioned East Asian faces. Interestingly, the generator sometimes added culturally specific clothing without any prompting, showcasing evident stereotypes, such as wearing a bindi or a sari. It also inexplicably blended distinct cultural attire, exemplified by a wedding image featuring a mix of qipao and kimono styles.
Sato remarked dryly, "Multiculturalism is amazing."
Age-related biases were also evident, as the bot frequently assigned older Asian men with surprisingly youthful Asian women, failing to accurately represent the diversity of marital relationships within the Asian community.
Reflecting Broader Societal Biases
AI models reflect not just the data they are trained on but also the biases of their developers. This was noted by many on Twitter, highlighting that mixed-race marriages are not a recent development; in the U.S., they have been legal for over 50 years since the landmark Loving v. Virginia Supreme Court case.
According to recent statistics, 17% of marriages in the U.S. involve partners of different races or ethnicities, with 29% of Asian newlyweds marrying outside their race. Ironically, Meta’s founder, Mark Zuckerberg, is also a white man married to an Asian woman, Priscilla Chan.
Sato points out that in Western media, “Asian” is often treated as a homogenous term, despite the continent’s vast cultural diversity. With nearly 5 billion people—60% of the global population—Asia consists of an array of distinct cultures.
She concludes, "Perhaps it’s not surprising that Meta’s system assumes all ‘Asian’ people look the same, when, in fact, we’re a diverse collection of people who often have little in common besides ticking the same census box."