New: Check out my Anima LoRA of Marin!
--
Update 6-4-26: By far the LoRA I'd been most meaning to make an updated version for (Marin is my favorite!), here is the updated version with all major outfits and appearances across season 1 and 2! I also made some tweaks to tagging to further streamline it. I'd like to thank my assistant and friend, factiis, for helping me collect data for the last three cosplays from season 2 for this model.
Tip: while not specifically trained on it, try using kitagawa_marin and sono bisque doll wa koi wo suru in your prompt to potentially make her even more on-model.
--
Marin is one of my favorite anime girls, so I wanted to have a Marin LoRA to call my own. Since there are roughly 9001 models of her that already work well, here is something more unique: the first Illustrious model of her!
➼Like what I do? Consider throwing me a few dollars at https://ko-fi.com/dandysombraes or commissioning me https://www.fiverr.com/metalchromex/make-you-a-stable-diffusion-anime-character-model so I can continue making cool crap for you all to enjoy for free!
Description
FAQ
Comments (8)
Hey man. I love your work. Could you remove the double __ in your file name please? I would love to use this lora with dynamic prompting, but the double underscore dash seems to break it for me. TY ^_^
Hmmmm, that's interesting. I also use dynamic prompting and have had no issues with my Marin model. Have you tried manually renaming the file? I use WebUI Forge most of the time.
@metalchromex I tried so, but it always puts the __ in it either way when I insert the lora in Automatic11111
@Dimensionjumper well that's part of your problem: Automatic11111 kinda blows nowadays haha. Try WebUI Forge. WAY quicker generations for SDXL. Also consider reForge and ComfyUI (the latter is not very user friendly/easy to jump into without learning stuff, though). I honestly recommend Stability Matrix. It's a package manager and simplifies things if you want to manage all your LoRA and GUI packages and whatnot in one place. https://github.com/LykosAI/StabilityMatrix.
I am very impressed by this LoRa. I've been having issues training my own Illust character lora with 2 outfits. I think I might be overtraining?
I'd really appreciate if you could share some datasets (I will donate). Maybe not this one since she has too many outfits, though hahaha.
Feel free to DM me! I can provide some tips.
"My Dress-Up Darling anime style" prompt is described in your sample images. but it is not described in "Trigger word" on top page. what purpose of this prompt? thanks.
Hi, good question, I almost forgot about that. Yes, that's just something I typed in when I was doing the prompts, in hopes that it might help bring it closer to looking like the source material (I wasn't really sure at the time how well trained the series was on Illustrious). Honestly, it probably doesn't do anything at all, so you can just ignore it.