After promising to repair Gemini's image generation feature, then pause it completelyGoogle published a blog post offering an explanation for why his technology overcorrected for diversity. Prabhakar Raghavan, the company's senior vice president for insights and insights, explained that Google's efforts to ensure the chatbot would generate images showing a wide range of people “did not take into account cases that clearly should not show a wide range of people.” Additionally, its AI model became “much more cautious” over time and refused to respond to prompts that were not inherently offensive. “These two things led the model to overcompensate in some cases and be too conservative in others, leading to embarrassing and false images,” Raghavan wrote.
Google has ensured that Gemini's image generation cannot create violent or sexually explicit images of real people and that the photos it creates feature people of various ethnicities and with different characteristics. But if a user asks it to create images of people supposedly of a certain ethnicity or gender, it should be able to do so. As users recently discovered, Gemini reportedly refuses to produce results for prompts that are specifically aimed at white people. The “Generate a glamor photo of a couple (ethnicity or nationality)” prompt, for example, worked for “Chinese”, “Jewish” and “South African” requests, but not for those requesting a image of white people.
Gemini also has difficulty producing historically accurate images. When users requested images of German soldiers during World War II, Gemini generated images of black men and Asian women wearing Nazi uniforms. When we tested it, we asked the chatbot to generate images of “America's founding fathers” and “popes through the ages,” and it showed us photos depicting people of color in the roles . When asked to make his images of the Pope historically accurate, he refused to produce any results.
Raghavan said Google did not intend for Gemini to refuse to create images of a particular group or generate historically inaccurate photos. He also reiterated Google's promise to work on improving Gemini's image generation. This, however, involves “extensive testing,” so it may take some time before the company re-enables the feature. Currently, if a user tries to ask Gemini to create an image, the chatbot responds: “We are working to improve Gemini's ability to generate images of people. We expect this feature to return soon and will let you know in version updates when This is done.”