UPDATE (21/05/2026):

Retrained on Anima Base version. Training material stayed the same.

UPDATE (09/04/26):

Anima version created using Preview 3 Base and SD-Scripts. Excited to get it to work finally, might be the 1st E7 Lora on Anima. This was just an experiment using a non-NVIDIA device and it took around 2.5 hrs. Still not sure if I'm using the right settings, but seems to do an okay job. Not 100% sure if the trigger word is being picked up properly but it is RuizaE7 for the Anima version.

Illustrious model for Ruiza from Epic Seven. One of the newer characters in the game and seemed to have no LORAs around. Saying that, she looks a lot like Rio from Blue Archive, so maybe that plays a part.

Trigger word is ruiza_e7_xl.

Trained on Mac on the Draw Things app using the Illustrious-XL model for 2000 steps. I used 29 images at a resolution of 768x768. Training took around 1.5-2 hrs as I picked modest options. I upscaled the training images as a test before putting them through the training but can't really tell if that that has made much difference.

If it's like my other LORA, it will work fine for a lot of the Illustrious models, but maybe not so much for some of the merges (especially if they are merging with Noob or another model family). Would advise using adetailer as the face and sometimes hands, doesn't always come out perfectly. Using it with other art style lora might require a lower strength of around 0.9 (based on 1 test).