Microsoft Reverts Bing Image Generator to Old DALL-E Model Due to User Dissatisfaction

According to reports, Microsoft has decided to roll back the AI model for its Bing Image Creator. In the weeks following the December 18 upgrade to the new DALL-E 3 model, numerous users complained that the tool's performance had significantly deteriorated.

Microsoft declined to comment on the rollback decision and did not provide detailed reasons for the discrepancies between user expectations and actual outputs.

Recently, Jordi Ribas, head of Microsoft's search business, tweeted that they had managed to reproduce some of the issues reported by users and decided to temporarily revert to an older version of the DALL-E model. However, this process could take several weeks to complete.

Since Ribas' tweet about the change in December, users have noted that images generated by Bing Image Creator lacked detail or failed to accurately reflect the content of their prompts. In his initial response, Ribas stated that the new model's output quality "should be better on average" compared to the previous version.

Similar complaints appeared on Reddit and the OpenAI community forum. On the OpenAI forum, users criticized the model's handling of anime-style characters, particularly noting issues with fabric textures. One post mentioned that while the left image was "perfect," the right one appeared "overexposed."

There were also complaints about Bing's added starburst effects.

These issues are subjective and do not definitively determine which image is better. It seems Microsoft now faces the challenge of addressing both technical concerns and the subjective critiques from AI art enthusiasts comparing machine-generated outputs to their envisioned creations. Perhaps seeking advice from the artists whose works were used to train the generator could help manage user expectations more effectively.