V4 continue leosam merge?

Something interesting: when training images with dreambooth i usually after merge it with a checkpoint to give the newly trained checkpoint more realism and to give it more versatility. I tried over 10 checkpoints but epicrealism seems to work best. Even better than Leosam's, juggernaut or realvis. The trained face stays sharp and although slightly changed, like with all merges, the most natural.