Because if you are talking about the GPU mode that's... weird. From what I saw, the version @balpha suggested - github.com/bes-dev/stable_diffusion.openvino - seemed far faster than this one when run in CPU mode.
Yet... I don't think that balpha version has an easy support for switching the actual model checkpoint (if for example you want to try the "Waifu Diffusion" model - very annoying name, I know... but the model is trained only on anime pictures, and that shows a lot if you try to generate anime style art)
(the first is an attempt at generating something looking like EG Fluttershy by actually describing the character. The second one is "Aerith in a kimono, sakura bloom background" and the usual modifiers like "8k extreme detail" )