<p>Same images, better captions, better training process, more training. Results now have less original works in them with roughly same amount of style, meaning that this LoRA now better responding to prompts and keeps more of checkpoint style in place. This also means, that if you use overtrained checkpoint with very strong style (like WAI-NSFW) you might want to increase strength a little. Or not, who am I to tell you what to do?</p><p>While this version a lot better than previous, it still introduces some odd effects, like rock borders, men in women's clothes, default character positioning in distance that mangles faces, etc.</p><p>Default strength of 1 is good. You can go as low as 0.3 or 0.4 (depending on checkpoint) for just subtle style hints. I don't recommend going above 1.3, but again, who am I to tell you what to do?</p><p>As always, will appreciate your pictures and comments that say what works for you and what not.</p>