Google's Gemini Fumbles Image Generation With Overly Strict Diversity Filters
Google faces backlash after its Gemini AI produces historically inaccurate images due to excessively strict internal diversity filters. The company temporarily disables the chatbot's ability to generate images of people while it investigates the issue.
Google experiences significant backlash this week after its Gemini AI chatbot produces historically inaccurate images of people. In response to complaints about the AI generating inappropriate depictions and refusing requests involving race, the search giant completely disables Gemini's ability to create images of any person, whether living, dead, or fictional.
The root cause of the problem stems from how Gemini interacts with Google's Imagen 2 image generator. Rather than generating pictures directly, Gemini acts as an intermediary that takes user prompts, applies its own interpretive filters, and passes the modified instructions along to the image creation model.
This middle-management process applies overly cautious diversity filters to all requests for people, resulting in bizarre and historically impossible outputs. Google states that it is currently investigating the issue and working to find a safe way to adjust these internal instructions so the AI functions as intended.