This lora was trained on 25-30+ images from what I <s>believe</s> know to be Himukai Yuuji's artwork. These are images I fount while searching for resources to train my Etrian Odyssey lora. I really liked the style so I attempted to recreate it with the small available resources. The good news is, I captured the style. The bad news is, (the pony version) is notoriously bad with hands and feet. Nonetheless, I like it.The Pony version was trained on <a target="_blank" rel="ugc" href="https://civarchive.com/models/288584?modelVersionId=324619">Autism Pony</a>. The illustrious version was trained on <a target="_blank" rel="ugc" href="https://civarchive.com/models/852957/zootlustrious-aesthetic-illustrious-xl-refinement">Zootlustrious V2</a>. The Noob version was trained on <a target="_blank" rel="ugc" href="https://civarchive.com/models/833294?modelVersionId=1190596">Noob 1.0 V-Pred</a>. The Anima version was trained on, well, <a target="_blank" rel="ugc" href="https://civarchive.com/models/2458426/anima?modelVersionId=2945208">Anima</a>. I haven't tested it with anything else.To get the results you see in the images above, <a target="_blank" rel="ugc" href="https://civarchive.com/articles/11427/my-current-sdxl-settings">these are the settings I used.</a>

Himukai Yuuji Style | Anima/SDXL

Trained on 25 images at 1792 resolution. I went back and re-selected and edited the images to make sure I captured the style I originally intended. After thorough testing and retraining a few times, I am now 100% confident you can faithfully recreate the intended style with this version of the lora. Funny enough, I couldn't find half of the images I used in the old dataset as I was searching online. (I wonder why...? lol) Luckily, I still have all of the original data, with the most important ones being in good quality.You might get some wrong hands and feet every once in a while. Just add 5 fingers or 5 toes to the prompt and it should fix itself.

Files with SHA256: 65d02f7fe06ce12b03b3f706b64abebae7e2b178afc6fb81fcbe4ece4c191829