X
Innovation

Why Google just banned Gemini from generating images of people

After less than a month in Gemini, Google's AI text-to-image generator is already in hot water.
Written by Sabrina Ortiz, Editor
Google Gemini logo on laptop screen
Maria Diaz/ZDNET

AI image generators allow you to generate images of anything you can think of, including historical figures. However, when users started asking Google's Gemini to depict historical figures or groups of people, they were in for a surprise. 

In the past week, when users asked Gemini to generate images of historical figures or people of different races or nationalities, they began to notice that none of the images were true to the prompt. Rather, the images failed to render white people, even if the individual was white. Instead, it only generated images of multicultural and racially diverse people. 

Also: How to use Gemini (formerly Google Bard): Everything you should know

Users began taking to social media to share these images, in which prompts that asked Gemini to create images of the Pope, Father of America, and a Viking, all resulted in images of people of color. Former Google employee, Debarghya Das, shared an X thread of these inaccuracies compiling examples from user posts across X, as well as his own experience. 

As a result, on Wednesday, Google Communications acknowledged the issue via X, stating that it is aware of the inaccuracies and was working on improving those depictions because the chatbot "missed the mark." 

Then on Thursday, Google shared that the company is pausing image generation of people in Gemini as it works to address the issues, and the company is planning to re-release an improved version soon. 

If you try to ask Gemini to generate a photo of a person now, you will be met with a series of different error messages, such as "That's not something I'm able to do yet," and it won't generate the image for you. 

Also: I just tried Google's ImageFX AI image generator, and I'm shocked at how good it is

Despite launching its AI chatbot in March of last year, Gemini has only had its image-generating capabilities since the beginning of this month. With that long wait before releasing its image generator to its chatbot, it is surprising that it is already making these missteps. 

Google had a similar experience when it first launched Gemini, then called Google Bard, and the chatbot shared inaccurate information about the James Webb Space Telescope in its demo. 

Editorial standards