This LoRA was one of my earliest, while I was learning training. I have decided to retrain it again, for strange reasons. SD1.5/FluffyRock versions are still faced with instability, but SDXL versions perform much better, particularly the NoobAI variant.
Activation keyword is nanigashi; LoRA weight should be around 0.9 ~ 1. If aiming for maskless variants, increased tag weight may be necessary for them.
Recommended positives: white coat, chest belt, yellow horns, yellow eyes, gauntlets, cuisses, sabaton, leg armor, 3 toes, 3 fingers (Use unworn mask, mask off, black sclera for XL models when mask is undesired)
Recommended negatives: (I cannot think of many applicable negatives, but consider anthro, feathered hands for SD1.5/FluffyRock models.)
Description
Version trained on NoobAI, mostly tested for zatochu's merges, such as "Zoinks" and "Jinkies." Dataset and training are mostly similar of the Pony Diffusion version, so prompts are similar. ??????? improvements are significant, compared to prior versions. May still be stubborn to certain poses.
(Preview imagery generated in wildcard batches, for time conservation. Clean-up and upscaling are exempt for this reason. Prompts may contain nonsense as a result, except for character details.)