This GLoRA attempts to ease generation of Gunsmith Cats-styled imagery without the drawbacks of Illustrious' native understanding. There is no required trigger word, nor is it necessary to use additional style tags such as 1990s (style), retro artstyle, etc.
Preview Image Settings: DPM++ 2M SDE, Karras, 20 Steps, CFG 5
TXT2IMG outputs were refined using IMG2IMG, all same settings except steps were increased to 50.
Description
FAQ
Comments (7)
Holy lord, this model is WELL MADE.
Please for the love of god, share your training data info and general settings with us mere mortals!
Civitai needs more Loras of this quality!!!!!!!
Thanks for writing but I can't take any credit haha. Training settings are unchanged from the config provided by justTNP's article, linked below. My dataset is simply a bunch of 1080p screenshots of the Gunsmith Cats anime, autotagged with WDv3Large and then stripped of style tags to avoid those becoming needed in prompting.
https://civitai.com/articles/680/fictional-characters-among-other-things-a-lora-training-guide
Works awesome for the style. If you will ever consider doing V2, might I suggest tagging characters as well? While the lora can create a specific character by accident, it has no idea who Minnie May is, for example.
Hey, thank you for the rating and suggestion. Training for specific characters is not the goal of this GLoRA, but I did attempt to recreate a couple for the first two preview images anyways just to make this upload stand out more in a search. I apologize for any confusion or disappointment.
@Irastris Oh no no, like I said, the style is absolutely spot on, and if characters are not the goal, then there's no problem whatsoever. Was marely a suggestion, noothing more. Fenomenal work there, mate :)
@KarotConCarne who cares about the characters! only important thing is: can it correctly replicate rally's cz-75??? huh, huh? :D
jk, thx for the model, looking great.
@eurotaku Considering the show, I'd say the guns are more characters than the characters themselves ;D