This is NOT my model, but a converted version of align_sd to be used in A1111 WebUI, the original version is in diffusers format.
It is claimed to be better aligned with user intentions and produces images with fewer artifacts, such as weird limbs and faces.
According to original author, please add trigger words "weird image" into NEGATIVE prompt for better performance.
In my testing, it can sometimes help improving overall quality, but how it performs with different base model and/or embeddings/loras is remained to be found. But remember that there is no silver bullet in AI.
Refer to original repo for more information and licence.