2023/4/6 update
Upload Lycoris version. v1.0 version is conventional LoRA model.
Introduction
A conventional LoRA model of KDA Series, including 4 characters, Ahri, Akali, Evelynn and Kai'sa. 0.6~0.8 weights are good. The preview images are generated using 1.6 version (original) YesMix.
Usage
The universal trigger tag is 'k/da \(league of legends\)'.
For Ahri, the trigger tag is 'ahri \(league of legends\)'. The key features are 'fox girl, fox ears, fox tail, blonde hair, choker, facial mark, whisker markings'.
For Akali, the trigger tag is 'akali'. The key features are 'baseball cap, choker, cropped jacket, ponytail, mask (optional), kunai (optional)'.
For Evelynn, the trigger tag is 'evelynn \(league of legends\)'. The key features are 'choker, crop top, feather trim, midriff, hair between eyes, purple hair, claws, sunglasses (optional)'.
For Kai'sa, the trigger tag is 'kai'sa'. The key features are 'choker, double bun, hair bun'.
See the preview images for details. Btw, this model is trained with some images that contain the baddest outfits and prestige outfits. Therefore, images that contain the above outfits can be also generated using this model, theoretically.
Cautions (Must Read)
Some preview images are generated using hires fix. Thus, if you wanna generate the preview images, pls download the original preview images and analyze them using 'PNG INFO' in webui to get all generating information.
The maximum of generating resolution should not exceed 1024 theoretically, otherwise extra/deformed head/limbs will be waiting for you :). Try to use 'hires fix' to prevent this problem and help to generate higher-resolution images.
For 784mb VAEs (NovelAI, Anythingv3, Orangemix, Counterfeit) users, make sure add '--no-half-vae' in command line of webui to prevent generating black images. See voldy's log for the details of edit method. For high-performance GPU, make sure 'medvram' and 'lowvram' mode are disabled.
Description
FAQ
Comments (11)
How do you train multiple characters in 1 Lora? Do you make it as separate folder ?
yeah, one character one folder.
@zakp Okay thank you and you put them inside the training folder?
@mariegold yeah
I suppose it's LoHa/Locon for dual/triple/etc. training without bleeding mixes. Not LORA, right?
Do you have an update planned for this?
It will be updated if the training strategy is updated. Because the dataset contains all published images about KDA Series.
Can u make lora for style of artist LAM?
looks good. i'll try about it when the new training strategy is finished.
@zakp can you make Lora for style of Granblue fantasy game?
gbf(.)wiki/SSR_Characters_List
@zakp thank u very much