Alber, commonly translated to “Fool” or “Fool’s Guard” in English, also known as Porta di Ferro (“Iron Gate”) in Italian traditions, is a sword fighting stance/guard common across various traditions of European longswordsmanship. The German name (and most common English name) is thought to come from how it looks open, but is actually in a prime position to counter. It is equivalent to Gedan-no-kamae (“lower stance”) in Japanese swordsmanship.Trigger: alber_guard Support: fighting stance, sword, two-handed sword, holding weapon, weapon, full body, from side, facing viewer Not sure if will help, but trained with: Right foot forward, left foot forward, from left, from right, longsword, bokken, federschwert (it’s a type of blunt practice sword, only use one sword type in positive) Negative: Photo (medium), helmet, fencing mask (Don't use it as a positive: PDXL doesn't appear to know "fencing mask" well and this only has 6 pictures of three different people tagged with that so it will break images), any sword type not in use Can combine with: <a rel="ugc" href="https://civarchive.com/models/433869?modelVersionId=483914">CitronLegacy's lightsaber LoRA</a>. <a rel="ugc" href="https://civarchive.com/models/305864/pony-and-xl-weapon-katana-by-hailoknight">HailoKnight's katana LoRA</a>. (there aren't actually a lot of models that say how to draw a particular type of two handed sword that aren't also locked to a pose)I saw there were a few other sword stances on the site and they mostly worked, so I knew PDXL could produce half decent sword poses with a style LoRA (SD hands aside), but they were for broad classifications of poses and/or just plays on popular media poses instead of anything really suitable for a fight. I chose Alber to start with because it’s very photogenic (not having the blade overlap with the body or covering the face with arms/blade) and relatively consistent (unlike the more common Pflug/Plow which has a lot of significant variations) and its is very close to its Japanese equivalent.I trained this on 28 images taken from various HEMA instructional videos and photos, plus 5 instructional images for kendo. I think its come out a bit hit and miss (and thus not enough to follow with other guards), but maybe the posts will show better results or figure out how to get it to work more consistently. Failures seem to often produce the messer/one handed equivalent despite never appearing in training data. Seems to have minor influence on default background due to most training data being filmed against a wall, but it's weak (easily overcome with more detailed background requests) and often desirable (over Pony's tendency to devolve into non-Euclidean scenes, where objects are simultaneously background and foreground, when it tries to make complex indoor scenes)

Files with SHA256: f5a9f656abd09ed88c296e5b08ae7a0c7c3c95acf0bc37fdcdd094cfe97dd033