Eina Tulle from Dungeon ni Deai wo Motomeru no wa Machigatteiru Darou ka aka Danmachi
V3 - Experimental LoRA make using another LoRA which made using Koikatsu which then was mixed with styles and quality prompts to break the Koikatsu style then images from this model used to train a new LoRA
The Swimsuit is depicted in 2 images in the dataset so may appear if you try to prompt for it but I didn't train the custom swimsuit as the goal here was just to see if I could get this to work without any additional fanfare
V2 - Better dataset tagging, more images added, ability to use quality tags now, outfit variation, custom swimsuit added from Koikatsu. I am actually impressed with how this turned out
She has no bust size trained into her so try out whichever one fits your fancy although "Large Breasts" seems to work okay for her office outfit
V1 -This started out as a test but I am somewhat happy with the results so thought I would upload it. There was some weirdness involving her sometimes generating holding an object when quality tags were used, if that happens try prompting against "Holding" I have no idea what caused things like riding crops and spatulas to appear but it was quite amusing
Description
This is something a bit different. I took 5 clear images of Eina and merged them with Koikatsu images so that I had a LoRA which would show Eina's face clearly but her outfits poorly, I then ran the Koikatsu images of her outfits through Img2Img and using a style LoRA with various quality tags I broke the Koikatsu style and upscaled the KK images to better quality. Then using the new higher quality images which showed both Eina and her outfits I retrained the LoRA
I don't know how this model will turn out if you use it as this was a crazy experiment I was trying for something else I wanted to do and the results seem promising enough
You'll probably want to use ADetailer but most people do anyways
FAQ
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.