V1 - AlternateRe-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 ReleaseFirst 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.

Marina Katsuragi - Kouyoku Senki ExS-TiA - Lusterise

Requires playing around with weights and tags, but you can mix and match transformations and clothing to some extent. No fine tuning of model (yet) and it could use it.

Files with SHA256: bad671ab4b7ef23aa28ad82753ea707936896a5f61b3e283af30029c6a26560f