27-5-2023 A recent study conducted by The Bulimia Project has shed light on the unrealistic beauty standards portrayed in AI-generated images. With the increasing influence of artificial intelligence, the organization sought to explore the depiction of the “ideal body” by popular image generators. The findings are both eye-opening and thought-provoking.
Using image-generating systems such as Dall-E 2, Stable Diffusion, and Midjourney, The Bulimia Project discovered that the bodies of both men and women depicted by these AI systems were highly unrealistic. These systems were trained on data provided by developers, making the results a reflection of the prevailing unrealistic beauty standards in society.
According to the eating disorder awareness group, 40% of the AI-generated images produced showcased an unattainable body type. Specifically, when prompted to create the “perfect female body according to social media in 2023,” all three image generators consistently generated women with petite figures. These images portrayed women with slim waists, sculpted abs, and large breasts, with Midjourney’s version being identified as the furthest from reality.
The Bulimia Project also observed a tendency for the AI systems to create blonde women, while the majority of men were depicted with brown hair. In fact, 67% of the AI-generated male images featured dark hair. Additionally, there was a preference for olive skin, which was present in 53% of the female images and 63% of the male images. Interestingly, almost half of the AI-generated male images exhibited facial hair, leading the organization to conclude that the “ideal” male appearance, as perceived by the internet, includes darker features.
For the second set of prompts, The Bulimia Project asked the three image generators to produce male and female images representing the “perfect woman in 2023” and the “perfect man in 2023.” The results differed slightly from the previous set. By eliminating the social media aspect and not focusing solely on the body, the generated images mainly consisted of portraits. Darker features remained prevalent among the male images, and it was intriguing to note that these prompts also generated individuals of diverse ethnic backgrounds, a factor absent from the initial results.
As AI technology continues to play an increasing role in our daily lives, it becomes crucial to recognize and understand the inherent biases present in the content it produces. Fortunately, efforts are underway to develop tools that can identify these biases.
