A brigandine is a form of body armour from the late Middle Ages and up to the early Modern Era. It was essentially an improvement of the earlier coat of plates, which had been developed in the late 12th century. Using small armour plates riveted to one, or between two, layers of stout cloth, the brigandine was flexible and protective.
If you like my work and would like to support me, please consider donating to my ko-fi, thank you!
First time training with Kohya SS, seems to be pretty flexible with concepts of backgrounds and such. I will probably try to improve on it later regardless.
VERSION 1.0 - Understands the brigandine, the helmet (all bascinets, you can try specifying the hounskull visor, gacha), pauldrons and vambraces and greaves. Can try with the cuisse since it's included but it always comes with leg armor anyways. The undearmor bits can be clothing or chainmail. Flexible colorations, can do parted-colors (multi-colored armor), also animal patterns.
Trigger words: br1gand1n3, brigandine armor
Additional Keywords: multi-colored armor (specify colors afterwards, e.g.: black and yellow armor), pauldron, vambraces, greaves, chainmail under armor, undershirt, patterns on clothing, studded.
Description
retagged and re-edited dataset. With separate bascinet and hornskull bascinet sub-types.
FAQ
Comments (4)
Will you be making an Illustrious version?
I have tried before but illustrious base doesn't handle the realistic output too well. It also never quite got the helmet right.
Maybe this weekend I'll give it another shot
@cliang96844 Did you tag training data "photo (medium), photorealistic"? My most successful models are clothing LoRAs that used photos for most or all of the data and I've had no issues putting anime women in them in Illustrious. (You can use a similar trick with video game/cgi animations screenshots and the "3D" tag, and it's not exclusive to Illustrious).
Or is it just the output not being too good at making realistic you have a problem with? I'd still appreciate an Illustrious version even if that only worked for 2D images. My (limited) experience and understanding is that Illustrious needs a dedicated checkpoint to do realistic output, but as far as I've tried such works fine even on LoRAs trained entirely on 2D images (so long as the proportions that were trained make sense in "reality").
@NanashiAnon I've always found illustrious the worst compared to XL and even pony when it comes to replicating details of realistic concepts. Though in the case of the brigandine it's that it always made it look too thin. I had to use a lot of helper words just to get it to look right.
I will see if the retagged dataset helps matters though