This is a LORA of the character Velvet from Vanillaware's Odin Sphere. To prompt this character, enter "<lora:Velvet-OdinSphere-v0.9:strength>velvet, red hood, crop top, detached sleeves, string ribbon, star hat ornament, star_\(symbol\), long skirt, vertical-striped stockings, high heels," This LORA is usually the best looking at 0.6 ~ 0.75 strength. This dataset was a nightmare to work with, a constant uphill battle. Sources online do a horrible job of being consistent with her stars on her hood, and the black lace cap. The big issue is that 75% of her outfit isnt easily identified by the correct definitions of her outfit. I still do not think this model is ready, but I have spent an absurdly long time trying to troubleshoot and make this work. Despite being trained NAI, it seems to work better on Anything/AOM2. Problems: The lace cap hat is a nightmare. Due to a number of inconsistencies in the best images I could find, the black lace cap comes out much, much worse when prompted at a full body distance, compared against upper body closeups, where it looks much better. I believe this is due to the quality of the cap in the dataset, which looks far worse at a distance in the art. If you know how to fix this problem, please let me know, cause this has been bothering me ever since I started this character. Also, maybe if you're unlucky she'll grow an extra limb. Funny shit, if you're lucky when you prompt her, you may perchance have her generated with a drink in her hand. If she has a drink unprompted, congrats! You win! |