A couple other people on here beat me to this one by a few days, but by the time I noticed the training was already running and it came out better than expected, so screw it, here you go. This LoRA attempts to capture the style of the Korean artist "ji_heeeeeee". <ul><li>Strength 1.0 works fine on models I tried.</li><li>No trigger word.</li><li>The artist specializes in solo shots of various characters, so the style excels at that. If you break out too much from that box it might struggle just a tad to maintain style coherency.</li><li>The dataset is all white backgrounds. So detailed backgrounds are a bit of a luck-of-the-draw situation depending on your model.</li><li>Adding "jaggy lines" may enhance the style a bit on some models.</li></ul>Have fun!

ji_heeeeeee - Style (Illustrious LoCon)

v1 seemed like maybe it was just a bit underbaked so I let it train for another few epochs. The change is subtle but I think it does slightly better backgrounds and adheres to the style better with fewer extraneous details.

Files with SHA256: 72819edef299ca04232b7cc738889666b0029a2347e30f7c42df30349205c942