EDIT: Added 1e4-05 in case you experience smeared/ruined eyes or some other overbake issue, added a few different comparison graphs to that version's preview (open in new window to view at full resolution). I think it doesn't follow the style so strongly so dunno if it's an improvement.Can be used as style model just fine, but also encompasses a fuckton of characters. First grid is misleading, it didn't actually have a chara lora loaded. The rest of Sumika ones are tested against her chara LoRA properly.I've squeezed every FMD Muramasa character in here. Around 20 characters. Almost all of them with shit datasets. Don't expect miracles. Well, it works, how good it is might depend on the character itself.Regarding versions I believe 1e4-10 is best, 5e4-10 is mostly the same but I think it's failed with the style(?), seems noticeable on the color pallette with Sumika. As for 7e5 I think it does the minor characters best, if you wanna try reproducing them. 14400 for archival purposes, might be useful for someone I guess.Anyways, since individual character LoRAs lost some artstyle in exchange for actually working and getting details right, this one exists to give them the official style back. Just use it alongside them - this one should probably have lower weight than the character one to not mess details up but you'll have to experiment.As for prompting for all the characters in here it's going to be part of their name in most cases. Your best bet is checking the tags via additional networks extension or trying either name or surname and seeing which works.Exceptions are "Muramasa Nisei" "Grandfather" and "Minato Subaru". Let me know if you want any of these characters separated into a proper LoRA. Though alot of them don't have enough images for proper datasets so might end up like Doushin or Kageaki.

FMD Muramasa

Muramasa_14400.safetensors

Files with SHA256: 9ef10d443ba1c7a3b31d3b456a9b7c8ed83e243abe472349bc6a9e93295e8da8