LoRA model of Chracters from Genshin Trained by myself
Brief Intro
All these models are trained basing on Stable Diffusion 1.5, or NovelAI 1.0. All these model are trained with both illustraions and 3D models with 1024x1024 resolution, to ensure accuracy and versatility. Detailed training settings could be found within the LoRA model files themself, so check it yourself if you are interested. All of the models here can be found and downloaded form my page on CivitAI with all the information. Here is a place for purity.
The Prompts
For most of my model, I am used to use the standrd danbooru tag format as the trigger prompt, thus save your time and efforts since you don`t need to write it down, or type them manually if you have the tag-auto-complete plug-in installed. For trained costumes, i.e. the default skin/costume of the character in the game, I am used to have a set of tags. On one hand, using these tags will guarantee the accuracy for generating the desired features of the costume; on the other hand, remove/replacing these tags makes it easier to remove/change the costume, i.e. good versatility. Sometimes, using my chracter lora without the trigger prompt, but with the set of tags of the costume, will make my character lora becoming into a costume lora.
The Weight
Typically all my LoRA models works with weight from 0.6~1, and 0.9 will be the recommanded value. However, lower the weight if you see too many distorsion, a sign of overfitting, or simply unremovable/unchangable clothings.