Generative artificial intelligence, or “gen-AI,” uses machine learning, neural networks, and other techniques to generate new content—like images, videos, and text—by analyzing patterns and information from massive training datasets. ChatGPT, Gemini, and MidJourney are just a few examples of popular services that use gen-AI.
It has been predicted that soon a majority of online content will be created by gen-AI. The speed and advancement of gen-AI can be impressive and present new opportunities. However, gen-AI also raises ethical issues that must be considered.
One of these ethical issues pertains to body image—how people think and feel about their body. While a lot is known about how “traditional” media imagery (using real humans) affects people’s body image, how imagery created by gen-AI affects body image is currently unclear.
To that end, my colleagues and I conducted an experiment to test the effects of two types of gen-AI imagery—idealized vs. diverse—on body image. Further, we investigated whether disclosure impacts these effects, and what appearance-based individual characteristics might determine who is more affected than others.
The Experiment
Participants were 291 women aged between 18 and 31 years. They completed questionnaires to measure their current body image (i.e., how appreciative and satisfied they felt with their bodies). Afterward, the participants completed the “gen-AI exposure.” Namely, all participants viewed 20 images that we had created using gen-AI, but the type of images they saw differed according to what group they had been randomly assigned to:
- Idealized: These gen-AI images depicted women with bodies that aligned with societal body ideals. Given that the “fit ideal” is currently the most popular body ideal, these women were lean and toned, and were shown engaging in physical activities such as running and weight lifting.
- Diverse: These gen-AI images depicted women with diverse bodies—for example, in terms of body shape and size. Moreover, these women were shown engaging in a variety of activities that exemplified the holistic nature of body functionality, beyond physical activities, such as creative activities (e.g., painting) and self-care (e.g., cooking). This decision was made based on our research showing that appreciating all of the diverse things that your body can do (rather than how it looks) is a powerful strategy to promote positive body image.
- Control: These gen-AI images were also visually appealing, but did not contain any people. They depicted relatively neutral objects such as towels and books.
In addition, half of the participants in each condition were informed that the images had been created by gen-AI, and the other half of the participants did not receive this message.
After the gen-AI exposure, all participants completed the same questionnaires about their current body image. They also completed questionnaires about their tendencies to compare their appearance to others’ and the extent to which they bought into societal body ideals (“internalization”). Individuals with higher appearance comparisons and internalization tend to be more vulnerable to the effects of media imagery; we wanted to test whether the same was the case when it comes to gen-AI imagery.
The Key Findings
Overall, we found that:
- Participants in the idealized group reported reductions in body image from before to after the gen-AI exposure, whereas participants in the diverse group reported increases in body image.
- Unexpectedly, we found increases in body image within the control group, who had viewed images of neutral yet visually appealing objects. (It is currently unclear why this is the case, though some research suggests that aesthetic experience can impact well-being, which may extend to body image.)
- The gen-AI images were more impactful for individuals with higher levels of appearance comparisons and internalization.
- Being informed that images were made by gen-AI was protective for some, but not all, aspects of body image.
- The effects of disclosure were especially beneficial for individuals with higher levels of appearance comparisons and internalization.
Take-Home Messages
To the best of our knowledge, this was the first experiment to test the effects of gen-AI imagery—idealized vs. diverse—on women’s body image, and whether effects differed when participants received a gen-AI disclosure.
Our findings suggest that viewing gen-AI imagery that depicts idealized bodies leads to poorer body image, whereas viewing gen-AI imagery that depicts more diverse bodies leads to improvements in body image. Thus, when it comes to gen-AI imagery of people, it is better to depict diversity compared to idealized bodies.
This is also the first experiment to create imagery that depicts diversity in terms of body functionality, beyond physical activities. In line with the broader research on positive body image, our findings show that promoting a holistic and appreciative focus on body functionality is beneficial for body image.
Body Image Essential Reads
Beyond these general effects, we found that more “vulnerable” participants were more affected by the gen-AI imagery: those with higher appearance comparison tendencies and higher internalization of beauty ideals. These findings may help to direct future intervention efforts to people who may need them the most.
Disclosure seemed to be beneficial for some aspects of body image that we assessed, but not for all. For example, even when participants who viewed the idealized gen-AI images received a disclosure, they still felt less satisfied with their bodies. It could be that they compared their own bodies with the idealized bodies in the gen-AI imagery, even though they knew they were “fake.” Appearance comparisons often happen automatically, outside of conscious awareness or control.
Last, our findings also suggest that disclosure can be especially beneficial for more vulnerable individuals—similar to the findings for image type. Thus, it may be worthwhile to explore the use of disclosure messages on gen-AI Imagery in the future.
As this is the first study to investigate the effects of idealized and diverse gen-AI imagery on body image, there are many questions that remain to be investigated. As gen-AI continues to rapidly develop and become more integrated into people’s daily lives, the urgency for this research will continue to rise. We hope that this experiment provides inspiration for future research in this area.
