An attempt to train a style LoRA based on my own drawings, in case you felt like your gens are missing that "amateur pencil drawing" feeling. Can't make any excuse about the quality of the data for this one I guess. I think this initial version is about as good as I can get with SD 1.5, will see if I can get a better SDXL version soon.Edit: SDXL version uploaded, while it's listed as for "NoobAI" it also works fairly well on other SDXL-based models (Illustrious, Pony, SeaArt, even base SDXL) as it was trained against multiple bases at different epochs (ex. training efhoscistyle-1 on Pony, then training efhoscistyle-2 on Illustrious while loading previous weights from efhoscistyle-1, etc). Went through way too many of those to tune it to where I wanted, not sure I fully got there but I think it's good enough for now. See example images for some of the models I tested it with. Using the DMD "distilled" LoRA seemed to work okay, though it often ignored the "black and white" or "grayscale" tags in the prompt. Can either be converted manually later or there may be a ComfyUI workflow that can help.See below for some links to my galleries if you want examples of the real thing. Content is mostly furry macro/micro and some male muscle and hyper. There may be some bias towards that in this model as well, though it doesn't seem to be overwhelming.<a target="_blank" rel="ugc" href="https://www.furaffinity.net/user/efhosci/">https://www.furaffinity.net/user/efhosci/</a><a target="_blank" rel="ugc" href="https://furrynetwork.com/efhosci/">https://furrynetwork.com/efhosci/</a><a target="_blank" rel="ugc" href="https://www.weasyl.com/~efhosci">https://www.weasyl.com/~efhosci</a>Use "efhosci style" for the general style, along with more specific style categories:<ul><li>pencil shading, greyscale</li><li>pencil outlines, black and white</li><li>digital colors (flat colors applied to a layer under a previously greyscale shaded pic)</li><li>colored pencil (colored using colored pencils from the start)</li></ul>Can end up with some anatomical oddities. Results may be a bit better by applying the LoRA only to later steps of the generation, or using it at a lower strength initially with a stronger refiner stage.

Files with SHA256: 0fb7378fa5a487b4bdd0c80e66e8de02196bf6deee5d74900a92a39c68c062f3