The character Yusa Emi from The Devil is a part-timer!. I wasn't quite happy with the other model on this site and my old multi-character hypernetwork has difficulties with the braided sidelock, either placing it on the wrong side or making both sidelocks braided.My initial attempts on a Lora didn't work out well, but turning off "flip augmentation" which I didn't realize was on and after adding regularization images it started working fairly well.Main trigger work is <code>yusa emi</code> and that should be sufficient to get the face and hairstyle. The non-braided sidelock can be a bit inconsistent, perhaps due to the manga letting it run over the shoulder. Adding <code>sidelocks</code> might make it more consistent, I haven't tested it so much.For her office worker outfit you can use <code>emi-office</code>. I have not trained for her hero look or the fast food outfit.<h2 id="heading-1093">Version 2.0 updates</h2>I included screenshots from the anime to include more outfits and Maou and Alas Ramus. Some of the outfits could use some more work.<h3 id="heading-1094">Outfits:</h3>Only for Yusa Emi. Note that the tags changed from V1.0<ul><li>yellow casual: Yellow shirt with light gray jeans</li><li>fullsuit: outdoors office lady outlook, need additional tags to work: pantyhose, pencil skirt</li><li>docodemo: Call center outfit, is a bit unstable</li><li>blue fashion: pink shirt, blue jacket, and light gray jeans. Not stable, jeans can end up too blue</li><li>swimwear: Flower pattern bikini. Does not generate the skirt</li></ul><h3 id="heading-1095">Characters:</h3><ul><li>yusa emi</li><li>maou sadao</li><li>alas ramus: Not tested too much but appears to work together with the yusa emi tag</li><li>emilia justina: Hero look with silver hair and red eyes. Not intentionally trained, but some are included in the dataset. Add armor tag and perhaps you can get something usable?</li></ul>

Yusa Emi - Hataraku Maou-sama!

Trained on screenshots from season 1 and 2 to include more outfits, plus maou and alas ramus

Files with SHA256: 9d35501f50c175e57fd049d2147faa8c3bef79db9ebe45394d35b12f82a4d9b7