
Microsoft is rolling back a model upgrade to its AI-powered Bing Image Creator, reports TechCrunch. The rollback occurred after numerous complaints from users indicating that the tool’s performance declined following the upgrade to the new version of the DALL-E 3 model on December 18th.
Microsoft has not provided specific details regarding the decision to revert the model or the underlying issues that may have caused the disparity between user expectations and the AI’s output.
On January 8, 2025, Microsoft’s head of search, Jordi Ribas, tweeted that they were able to reproduce “some of the issues reported” and are reverting to an older version of the DALL-E model for the time being, although the full process could take several weeks.
Since the launch of Bing Image Creator last spring, users have generated billions of images with text prompts. I’m pleased to share our latest updates to enhance your creative experience. Starting today, we’re rolling out the latest DALL-E 3 model PR16, which will create images… pic.twitter.com/3p9HsYMlr6
— Jordi Ribas (@JordiRib1) December 18, 2024
Thanks again for the feedback and patience. We’ve been able to repro some of the issues reported and plan to revert to PR13 until we can fix them. All Pro users and about 25% of the requests using boosts are now on PR13. The deployment process is very slow unfortunately. It…
— Jordi Ribas (@JordiRib1) January 8, 2025
Immediately following Ribas’s announcement about the changes in December, a surge of complaints emerged from users who found that Bing Image Creator was generating less detailed results or images that did not accurately reflect their prompts. Initially, Ribas responded by asserting that the model’s output quality “should be a bit better on average” than before.
Complaints were also prevalent in discussions on platforms such as Reddit and OpenAI’s community forums. On the forums, users expressed dissatisfaction regarding the model’s performance with certain elements, including a complaint about the handling of fabric on an anime-style character’s dress. One user indicated that the image on the left was “perfect quality,” while the one on the right was described as “over-lit.”
/cdn.vox-cdn.com/uploads/chorus_asset/file/25823345/Screenshot_2025_01_08_at_5.49.25_PM.png)
Screenshot: OpenAI community forum
Another user criticized the way Bing placed starburst effects.
/cdn.vox-cdn.com/uploads/chorus_asset/file/25823341/Screenshot_2025_01_08_at_5.41.09_PM.png)
Screenshot: OpenAI community forum
These issues are inherently subjective, making it challenging to determine which output is superior. This situation highlights that Microsoft now faces not only technical complaints but also critiques from AI art enthusiasts, who have their own standards and expectations for the generated images. Perhaps seeking advice from artists whose works contributed to the training data could help Microsoft better manage user expectations regarding AI-generated content.