V1 - Alternate
Re-release of the LoRA with a new training dataset, as I lost the previous one. Put together pretty quickly and did not prune tags so there is a lot of bleeding between the trigger words. It helps to play around by adding the other trigger words to negatives and being detailed in your prompt if you want to try and replicate the outfits. Other than that, it is a bit cleaner than the first release since I cleaned up the images I used for training this time. V2 is coming eventually, and I will fix the bleeding between the different forms and make it so the outfits reliably generate based on the keywords.
V1 - Initial Release
First time attempt at training a LoRA. This one is based on Marina Katsuragi from EXS-TIA by Lusterise. It is trained on a complete set of CG and can replicate all of her major transformations. The sample images were created with Hi-Res Fix and SD Ultimate Upscale but were otherwise created solely through txt-to-img.
This can be used on its own with a weight of 1.0 and CFG of 7.0. 20 steps with DPM++ 2M Karas is sufficient. I hope to update it later once I accumulate some buzz.
Description
Recreation of V1 as I lost my training dataset and had to create a new one. Added some images into the training dataset to generate a fox girl form.
This LoRA generates differently from V1 so I figured I would post it for anyone who wants to use it. I'm putting some of the keywords in the trigger words section. They do not accurately prompt outfits though, so you will need to be detailed in your prompt.