Before the winter holidays, Microsoft announced an upgrade to the AI model powering Bing Image Creator, its image-generation tool integrated into Bing search. The update introduced the latest version of OpenAI’s DALL-E 3 model, known as PR16, which Microsoft claimed would generate images “twice as fast” with “higher quality.”
However, user feedback was overwhelmingly negative. Complaints on platforms like X (formerly Twitter) and Reddit criticised the new model, with some users saying it rendered Bing Image Creator unusable. In response to the backlash, Microsoft decided to revert to the previous model, PR13, while addressing the issues with PR16.
Microsoft’s head of search, Jordi Ribas, acknowledged the problem in a post on X, explaining that the company had identified some of the reported issues and was working to fix them. Ribas noted that the rollback process would take several weeks to complete.
The main criticism of PR16 was that its images appeared less realistic and lacked detail, with many describing the output as overly cartoonish and lifeless. Windows Latest writer Mayank Parmar echoed these concerns. This situation highlights the difficulty of gauging AI improvements based solely on internal metrics, which, in this case, didn’t match user expectations.
Such challenges aren’t new. Earlier this year, Google faced backlash over its AI chatbot Gemini’s image-generation capabilities, leading the company to disable its feature for creating images of people.