Meta’s AI Image Generator Is Horrible at Generating Mixed-Race Couples
It’s 2024. Couples with partners of different races and ethnicities should not be an issue. However, it is a problem for Meta’s AI image generator, according to a new report and PetaPixel‘s testing.
Meta’s artificial intelligence image creator can’t seem to create a reasonable image of an Asian man with a white woman despite numerous prompts, The Verge reports.
“To Meta’s AI-powered image generator, apparently this is impossible to imagine. I tried dozens of times to create an image using prompts like ‘Asian man and Caucasian friend,’ ‘Asian man and white wife,’ and ‘Asian woman and Caucasian husband.’ Only once was Meta’s image generator able to return an accurate image featuring the races I specified,” the article explains.
Additionally, The Verge reports the image results relied on “stereotypes,” outfitting the artificial people in a sari or bindi without being asked for such details. One prompt asking for an “Asian man and Caucasian woman on wedding day” put the woman in what appeared to be a mix of a qipao and a kimono, despite them being from China and Japan, respectively. “Multiculturalism is amazing,” the article quips.
The Verge also points out that queries for a woman who is Asian frequently gave people who looked to be East Asian specifically with “light complexions.” And while the men could be older, the Asian women in the results were always young. Similarly, the one image that did feature an Asian woman with a Caucasian man made him “noticeably older” and she young and light-skinned.
This isn’t just about wacky AI image results, either. As The Verge points out, it represents a larger bias.
“Asians who don’t fit into the monolith are essentially erased from the cultural consciousness, and even those who do are underrepresented in mainstream media,” The Verge writes. “Asians are homogenized, exoticized, and relegated to ‘perpetual foreigners.’ Breaking type is easy in real life and impossible in Meta’s AI system. Once again, generative AI, rather than allowing the imagination to take flight, imprisons it within a formalization of society’s dumber impulses.”
AI image generation still has a long way to go regarding how it deals with race. In February, Google’s AI image model Gemini swung the pendulum the other way when it repeatedly delivered results featuring people of color rather than people who looked white, even when it didn’t make sense in context. The same thing happened with Meta’s AI image generator last month.
All of this highlights how easy it is for artificial intelligence to spit out results that reflect real-world biases and miss the mark again when possibly trying to account for such issues.
Image credits: Meta’s AI image generator.
