It is my personal opinion. Training a checkpoint might not be a good idea. While I generate pictures, i don't want to mix concepts up. I don't want every harakiri girl with a noose on neck, neither want every hanged girl dissected. Specific loras are more flex than a compilative checkpoint. The checkpoint's quality is not as good as the base module WAI, and I can't use iLustmix or other modules to draw other styles. The last and the fatal one, the checkpoint affact face too much. That side-effect requires high weight of other character loras to override that may cause distortion.