





(I struggled to train the previous LoRA (which Derrian trained and uploaded it for me when I gave my dataset of Hatakaze from Azur Lane to him) and this one due to latest version of both Kohya and Derrian's scripts not working properly for me on Linux since they either give me a Nansexception error in WebUI when I use it because its loss value increases to then become NaN, or it doesn't nail the character right at all, so I had to get older commits of both scripts from GitHub. not gonna update 'em EVER AGAIN...)
Days ago, when I was browsing for LoRAs here, I came across one of Kuro/MDR from Girls Frontline: Project Neural Cloud, unfortunately it wasn't very good since it wasn't flexible at all and was missing one of her costumes being her halloween one called 'Spirit Trap'. In order to retcon this I decided to train my own LoRA of that sweet live streaming thot doll, this time including her halloween costume and making it completely flexible too. I gotta say that I'm really satisfied with the results, I approaced the method of using a main trigger tag for the character and a second for each of her costumes, I've only pruned her body and tags so that she can be flexible enough with any kind of clothing you want to prompt for her (more or less...)
Here's all prompts to generate her and each official costumes:
If you want her in any clothing you like, lower the strength to around 0.4~0.6 and put all of her costume trigger tags and some from each prompt above in negative prompt.
Trained with NAI as always, this time having used Kohya GUI from bmaltais in order to automatically guess epoch values from each folders of Kuro's dataset, I only used 1 epoch since I have an RTX 3060 and it took me an hour and 15 minutes to train her. Sadly it struggles at recreating each of her outfits correctly (except for MDR), since they're all quite complexly designed and very busy looking.
1. Modèle partagé uniquement à l'apprentissage et au partage. Droits d'auteur et interprétation finale réservés à l'auteur original.
2. Auteur souhaitant revendiquer le modèle : Contactez officiellement SeaArt AI pour l'authentification. Nous protégeons les droits de chaque auteur. Cliquer pour revendiquer
