A LoRA mainly trained on abdomera's Nasus custom skin and a few photos of cosplays ((an experiment to fix a messed up face) so, if your model tries to pursue realism but you don't want that, put "realistic" into negative.It was trained on NAI model, but well, it works better with semi-realistic models like JAM (tested on 1.9 &amp; 2.0, seems to work well with VAE (kl-f8-anime2), without VAE and with/without high-res fix (hit or miss though)). If you're using this lora on a nsfw model - changing chest size for some unknown reason would result bare chest (the dataset had no nudity).Due to a fact that 75% of dataset are 3D renders with not exactly the best textures, it usually goes for League-like 3d.I recommend using highresfix (r-esrgan 4x+ anime6b etc) with denoising strength of 0.5.It seems to mess up with the staff (if you want it, then at least put "multiple staffs" as a negative) and sometimes messes up the face. Also, original chest piece seems to be terribly inconsistent and the lora, overall, seems to be better with portraits.The model (custom skin) by abdomera.

Rule 63 Nasus

2E 2,000 steps, seems to have less weight towards 3D.works well with every model that works with 4E version

Files with SHA256: 0660b3c254b3f0492b530fa105d7e18fcff90c9991bc62280767cb9d8891135e