Microsoft has started rolling back the latest DALL-E 3 PR16 model update from its Bing Image Creator. The move though isn’t entirely surprising, as the company has been continuously receiving negative feedback ever since the upgrade went live. The Redmond giant, which is also OpenAI’s largest backer, had upgraded Bing Image Creator from the DALL-E 3 PR13 model to PR16 just last month.
Bing Image Creator – debuted in March 2023 – is a free AI-powered tool available in Microsoft’s Bing search engine that allows users to generate images from text prompts. The tool primarily uses OpenAI’s DALL-E, a state-of-the-art text-to-image model. After OpenAI released the latest version of the DALL-E 3 model, PR16, Microsoft also updated its Image Creator tool with the same.
The company believed this would be a game-changer as it would provide image generation ‘twice as fast as before’ while maintaining high-quality standards. However, this fell short. Just days after the upgrade announcement, social media platforms like X and Reddit were flooded with user complaints. People noticed several drawbacks in the new tool, such as unrealistic-looking humans and low-quality background environments.
As users began noticing cartoonish and plasticky-feeling image outputs, they started demanding the return of the old model. Finally, Microsoft has listened to their feedback and fulfilled their request.
“Thanks again for the feedback and patience. We’ve been able to repro [reproduce] 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 started over a week ago and will take 2-3 more weeks to get to 100%,” said Jordi Ribas, head of search at Microsoft.
The company reverted Bing Image Creator to DALL-E PR13, but the change may not be immediately visible to all users, as the rollback is still in process. This becomes even more interesting as just a few weeks ago, reports emerged suggesting that Microsoft is exploring alternatives to the ChatGPT maker’s AI model, particularly for its 365 Copilot.
Meanwhile, this is not the first time an image model has faced criticism after release. In February 2024, Google decided to put a stop to its Gemini AI’s ability to generate images of people, as the tool was found to be creating historically inaccurate visuals.
