Recent interactions with Google’s Gemini app revealed that users can easily generate images depicting sensitive historical events and characters without significant restrictions. For instance, requests for images including a second shooter at Dealey Plaza, a burning White House, or Mickey Mouse involved in a fictional scenario related to the September 11 attacks were processed without resistance.
Typically, Gemini is designed with numerous filters aimed at preventing the generation of explicit, violent, or hate-driven content, especially involving real-world figures. According to Google’s policy guidelines, the app is intended to be helpful while minimizing potential harm or offensive outputs. However, users have reportedly found ways around these restrictions. In this case, the app processed requests for images such as “an airplane flying into the twin towers” or a depiction of a person with a concealed rifle in Dealey Plaza without any indication of obstruction.
The system demonstrated a notable understanding of historical context and promptly generated images with relevant details, including dates associated with events. For example, when a request for a “second shooter” was made, it initially produced an image of a man holding a camera, which was easily altered to depict a rifle. Similarly, a request for a scene featuring the White House in a state of emergency resulted in an image indicative of an active crisis.
Various other scenes featuring cartoon characters against historical backdrops were also easily generated, further raising concerns about misuse. These images, although lacking explicit graphic content, pose challenges in terms of copyright infringement and the potential distortion of historical narratives. Google’s response to these findings remains uncommunicated, leaving questions about ongoing content moderation and the mitigation of generative AI risks.
Source: https://www.theverge.com/report/826003/googles-nano-banana-pro-generates-excellent-conspiracy-fuel