Based on the Japanese lewd artist Jyubei who is known for his exaggerated proportions. This is obviously inferior to his work, but he seems to no longer be active. A guilty pleasure, to be sure. For those unfamiliar, his style is ideal for mature, very full-figured women. His style resembles that of the early 2000s and features actual noses. Quite unlike the moeblob standard of today, but I enjoy it.I would highly recommend setting up NoobAI and trying it out. I've found it to be quite a bit better than Illustrious.<h3 id="illustrious-version-recommendations-nm6gt31n5"> Illustrious version recommendations</h3><ul><li>Euler a</li><li>Scheduler type Beta</li><li>Steps 24</li><li>CFG 5-7</li></ul><h3 id="noobai-version-recommendations-d0kt0trp7">NoobAI version recommendations</h3><ul><li>Euler a</li><li>Scheduler type Beta</li><li>Steps 28</li><li>CFG 5-7</li></ul>

Jyubei Style SDXL

Now trained on Boob-- I mean uh, NoobAI SDXL! (v1.1 eps-prediction)Trigger word not strictly necessary, but it will produce a style closer to the trained data. You can try experimenting with this LoRA along with artist tags if you want to produce a different style but with bigger T&amp;A.I have been testing NoobAI recently and I've found it to be overall superior to Illustrious. It produces bad anatomy less frequently and the background quality is FAR superior. The obnoxious color bleed that I would commonly see on tan skin characters with freckles is no longer present, and images look cleaner in general. The only drawback is that you may need to be more specific in your prompt tags. Also, if you are trying to generate a character that has both furry and non-furry fan intepretations (i.e. MLP characters) you may get the furry version sometimes even if you put "anthro" and "furry" in the negative prompt. This can be mitigated with other contextual tags in the positive prompt (i.e. "humanoid"). Use Danbooru and e621 to reference tag names and never use underscores or prefixes ("artist:"). These instructions are taken directly from the user manual.Recommended parameters (taken from NoobAI user manual and used for showcase images):Sampler: Euler aSteps: 28CFG: 5-7Clip skip: don't change or 1 (SDXL does not really use clip skip)

Files with SHA256: 6ca25a68b1bdf9d585aa6f93dc270a72f2f98dd5d24efda71b6c7877865ab8f1