It will do both Anthro and Feral(quadrupedal) versions of Chrysalis using feral or anthro in the prompt.Trained on both SFW and NSFW images, use rating:safe or rating:explicit in the prompt to force sfw or nsfw respectively. Trained at 512x512 resolution, but seems to produce decent results with different resolution and aspect ratios for checkpoints that support them (ex: 512x768). use Lora weight from 0.6 &lt;-&gt; 1 A weight of 1 may produce results with too much contrast. I use <a target="_blank" rel="ugc" href="https://civarchive.com/models/100191/ti-easynegativev2-textual-inversion-embedding">EasyNegativeV2</a> and <a target="_blank" rel="ugc" href="https://civarchive.com/models/87781/boringe621-negative-embedding-enhance-images-stylistically-and-topically">boring_e621_v4 </a>in the negative promptworks well with most stylized models. at least I think it was trained at 512x512, how bucket resolutions and "max_resolution" interact with each other (if at all) isn't really described or documented in Kohya. Still learning the ropes to LORAS, I usually just train hypernets.

Queen Chrysalis

Trained on 512*512118 images, 1 repeat8000 steps128 model and alpha dimension

Files with SHA256: 8db2435fdfaa885c9644a01b372fd707a94b8b5211b614baab717c0fc9979129