Changes for v1.6 : In short, does the same thing, but better - more data, more stable with prompts.
A LoRa trained on some concept arts of girls with weapons (and not only).
Although the name says "Girls with Weapons," this LoRa is not aimed solely at that. Ideally, the idea was to create a kind of constructor to satisfy my desire to create concepts, stuff in guns, a specific clothing style, and small details that complement the images and make them more original. Therefore, generating weapons is not mandatory if you don't want to see them.
This is only a test version, so it might work poorly with other LoRas and some models. So far, I have tested it on the base Illustrious v15, v14, v11, and Uncanny Valley. I am not sure about compatibility with other models, LoRas, and styles.
So, the key points of the first version:
Main trigger prompt: err0rf
1) Weapon generation with basic prompts like: weapon, holding weapon, weapon on back, katana, long katana, sword, giant sword, giant axe, giant hammer, rifle, giant gun, scythe...
(Probably sword
and katana
work best due to more data available for them).
2) Clothing style has a general prompt clothes details
, and prompts like: long skirt, short skirt, jacket, long jacket, long sleeves, hood, fake ears, thighhighs, armor, coat, dress, high heels, boots, thighhighs boots, mask, belts, gloves...
(Overall, it's a combination of traditional clothing and more modern styles like "techwear." Some prompts override those present in the checkpoints, so results may vary).
3) Cyborg body: cyber body, cyber parts, cyber legs, robot parts, cyber mask, mouth mask, fake ears...
4) Ears and tails: general prompts animal ears
and tail
. There isn't as much variety for ears, usually fox, cat, tiger...
The same goes for tails, including dragon tail
, as well as horns
.
5) Most of the concepts have a simple background
-white or black, or a gradient-so the background is not affected by the LoRa. You can add regular prompts for it.
It's important to note that most of the data has the full body
prompt, so it will generate a full-body character, as in concept art, unless specified otherwise. If you don't want a full-body picture, you can add prompts like cowboy shot, upper body, portrait
, also add full body
to negative prompts and remove prompts like sneakers
and similar to prevent the AI from wanting to create the lower body. Although I think it's clear what "Concept" in the name is for. Also For generating full body images, it is recommended to use a face fix tool like "ADetailer".
Generation settings I usually use:
Resolution: 832x1216 and 832x1312
CFG Scale: 2-5.5
Samplers: Euler A or DPM++ 2M SDE Karras.
You can generally look at the pictures I published for reference.