Unlimited Blade Works (UBW) | Weapon LyCORIS
V2:
Refer to the version details for Changelog and Triggers.
Weight can go from 0.7 up to 1. I recommend 0.9. ControlNet will streamline full body shots much easier, or more complex poses. Works with almost every Style, every Character and Photorealistic Models. I also recommend using some ControlNet Depth/HED for easier use.
Unlimited Blade Works (UBW) is now split into multiple individual models. Thanks to @Anzhc, the workload has become much heavier than before, due to his categorization tool, but it also yields much more consistent results. As a result, UBW will now receive 30 small updates/concepts over the course of each week. Some Concepts will receive multiple small updates to correct past mistakes. In the end, all models will be retrained as one large model, similar to the traditional UBW.
Everything below is V1.6 and Beta releases
Weight can go from 0.5 up to 1.0. I recommend 0.9. ControlNet will streamline full body shots much easier, or more complex poses. Works with almost every Style, every Character and Photorealistic Models. I also recommend using some ControlNet Depth/HED for easier use. I personally dont use them, but every help is appreciated. Some of @antonio_riolo2610 Models: v1 / v2. Also a big thanks to @Temporarium for helping me out with the Dataset.
Refer to the version details for Changelog. Keep in mind, that despite this LoCON, that Weapons are very difficult to generate. This only helps, but dont solve the issue.
This is now an official release of this Model and includes 23 CONCEPTS. Trained on 6,5k Images with 16,5k Training Steps. Trained on 7k Images with 45k Training Steps. Some of these are still unstable, even going over every Image manually. I also included some stances and Weapons that can be combined (Dual Wield of Axe, etc). Try some crazy stuff, even I couldn't fully test this Model after many hours. Here are the Triggerwords and helping prompts:
TRIGGERS:
Axe²
BowAiming³
BowHolding³
DualWielding²
Fan²
KatanaOneHand
KatanaTwoHands³
Knife²
Kunai²³
Lightsaber²
Unlimited Blade Works²
MasterSword³
PlantedSword
Scythe
Sheathed
Shield²
Spear²
Staff
SwordAndSheath³
Tomahawk²
Trident²
Unsheathing³
Warhammer²
Whip
² = can be combined
³ = very unstable
Helping Prompt:
holding, weapon, holding weapon, holding (Triggertag)
Stances Prompt:
Weapon over shoulder, Weapon on back, fighting stance, stance
BETA RELEASES:
This is only a BETA. I will add more Swords, Weapons and Guns to this. SD doesn't like to generate Weapons (just like Ranni with 4 Arms). Trying to improve it on every Update. I recommend using these Settings: Sampling Method DPM++ SDE Karras with 30/40 Steps. If you know how to improve this, i'm happy to hear you out. WARNING: This LoHA does have some problems and hits most of the time 1 out of 10 times. Hands are a big problem. Hand does not connect with the sword and sword does not appear in the right place sometimes. Unsheathing Sword is currently very unstable. Use pointy ears in Neg, if you are using Link Master Sword. Yor Knife gives normals knifes for some reason on my End.
Trigger Words:
Link Master Sword, Lightsaber, Dual Wielding Swords, One Handed Sword, Unsheathing Sword, Scythe, Sword and Shield, Staff, Trident, Planted Sword on Hand, Yor Knife, Unlimited Blade Works, Bow with Arrow.
I want to add these Concepts in the future as well:
Weapon Classes:
Axe, firearms, arc, Knife, hammer, rods, throwing stars, sickle, whips, guns, grenades, shields
Subcategories:
one-handed, Two handed, pistols, guns, machine guns, shotguns, grenade launcher, rocket launcher, sniper rifles, brass knuckles, throwing axes, spears, javelins, arrows, crossbows, Sling, sniper, bazooka, Foal
Variations:
warhammer, tomahawk, short sword, broadsword, rapiers, longsword, dagger, Kukri, katana, tanto, wakizashi, naginata, quarterstaff, war spear, double axe, hatchet, baton, Shield with integrated weapon, Dual wielding of other weapons
How to use LyCORIS's in auto1111:
Download LyCORIS
Copy the file to
stable-diffusion-webui/models/loraSelect your LyCORIS and use the triggerword
Make sure to change the weight if needed (by default it's
:1)Have Fun!
Description
Improved overall Dataset, increased Steps and higher chance. Switched from LoHA to LOCON.
FAQ
Comments (18)
Now this is pwetty ebic.
Have you thought about training at higher res and seeing how that goes? I realize you don't need the model to learn all the intricate details as that's presumably handled by the base model on upscale, but I'm kinda curious to see how it would turn out with such a large and varied dataset. I think it might be interesting if there's improvement, or It might actually be counterproductive if the downscaling actually helps unify some visual concepts that would differ in the smaller details.
Regardless, good job
Training on 512 does take a very long time, around 10 hours. Training on 768 or even 1024 will increase a little bit of quality, but that will cost me too much Gpu Power. I gain nothing from it while spending too much into it.
I was wondering, would you consider the next version having some colossal weapons? or at least a greatsword? Ive been wanting to try making a picture of Guts from berserk and using this to help it
I can try. For now, u could try plantedsword for similar effect
Any chance for sdxl version?
Planning after my Vacation
Hey there, well im currently doing some SDXL LoRAs, but its seems like, making this model (UBW) into SDXL will cost me too much. I calculated the process and it will lit take me 70+ hours of computing time to train this model for SDXL and I didn't even include upscaling the dataset + testing afterwards (probably another 30+ hours). Sry buddy.
@FallenIncursio Too bad, thanks for the update though :)
@FallenIncursio you'd rather train an XL checkpoint model focused on this material I guess. might cost the same and results may improve, as you will teach the checkpoint your own terms (tags, tokens) as well as publish the model everywhere in order to try to get some cash as revenue or prizes in return (since your dataset is huge and it's expensive to train it).
I install Lycoris, the perfect way, and it still doesnt work. What makes lycoris better than Lora?
Lycoris is a neural network that is structured differently than LoRA
Information is not stored the same way. In result, outputs are vastly different
The structure of the Lycoris network allows it to do things that LoRA can't do.
they say it gives more versatility, best quality outputs, but depends on how it was trained. i tried some and somewhat failed - still prefer my LoRAs instead. I need to read/learn/understand such process better, than maybe. seems also the file size is smaller... but a LoRA can be small too. But as Ive said, I need to understand better the process.
Please consider adding folding chairs in a future update, thank you
Well, time to gather Dataset for that
50% of it is for the meme but it's not like folding chairs don't have a history https://www.youtube.com/watch?v=AxldOEOnzPA
this is practically a checkpoint! i remember using it on webui, not sure if it was models, but all my men tend to wear female clothes, dress, skirt, even breasts lol. anyways, i love the series, nice job!
Is it possible to have a giant scythe?
Details
Files
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.


















