k-rha's style.<h3 id="v2-r3sfth76w">V2</h3>-Same dataset but new training settings and I refined the captions. -Added a trigger word: krhastyle-full sized model instead of a pruned oneAs for the preview images, in v1 I used base ponyXL and on v2 I used AutismMix Confetti. That alone makes a massive difference. But I feel v2 has cleaner more consistent results. It's more likely to gen illustration style concepts. If that's something you want, be vague with tagging backgrounds/environments. Tagging english text and border help push it in that direction a bit too.<h3 id="v1-zynkreo0s">V1</h3>For some reason, red eyes are more detailed. Likewise, pupils disappear sometimes. Feel free to add no pupils in your negative prompt.

k-rha (Artist Style) [PonyXL]

revised captions, new training settings, added trigger word "krhastyle", not resized.

Files with SHA256: 123b1bd62b8c223ea0ebf081c67c77f93a4653ee383b037b04f9e567562e812c