Microsoft has rolled back an update to its Bing Image Creator after users complained about a decrease in image quality. The update, which integrated OpenAI’s Dall-E model PR16, was initially touted as being “twice as fast as before and with higher quality.” Jordi Ribas, Microsoft’s head of Search, had previously announced the rollout on X (formerly Twitter) in December 2024, highlighting the “enhanced creative experience” PR16 offered.
Robot holding a video camera, generated by Bing.
This optimism, however, was quickly met with user dissatisfaction. Many users felt the image quality had significantly declined. One user on the Bing subreddit even declared, “The DALLE [sic] we used to love, is gone forever,” expressing the widespread disappointment. The sentiment was further fueled by comparisons of images generated by the previous model, PR13, and the new PR16, with many users finding the latter’s output inferior.
pre and post Bing update image quality comparision
Responding to the negative feedback, Ribas announced on X that Microsoft would revert Bing’s image generation model back to PR13. He acknowledged that Microsoft had been able to reproduce some of the reported issues and stated the company’s commitment to resolving them before redeploying PR16. Unfortunately, the reversion process is expected to take several weeks, potentially up to three, to complete fully.
This incident is not an isolated case in the AI image generation landscape. In February 2024, Google temporarily disabled Gemini’s image model due to the generation of racially offensive imagery. This highlights the ongoing challenges and complexities of developing and deploying AI image generation technology responsibly.
In conclusion, while the pursuit of faster and higher-quality AI image generation is a commendable goal, user experience and ensuring responsible AI outputs remain paramount. Microsoft’s decision to revert to the previous model demonstrates a commitment to addressing user concerns and prioritizing image quality. The ongoing development of AI image generation technology continues to be a dynamic process, marked by both advancements and necessary adjustments based on real-world feedback.
