Sol Marron - Black Clover [PonyXL]

The cute Sol Marron with a little more brawns than brains from the Black Clover series.How to useUse trigger word "slmr" and then prompt away. Her full default outfit is tagged as "crop top, blue capelet, white shorts, white shoes, belt". I recommend adding "midriff" and "cleavage" for best results. "bare shoulders" also work well with the capelet removed. Her physical appearance tags are pruned and doesn't need prompt but I found adding "green eyes" made them pop out more.A word of disclaimer - it was very hard finding good images so the dataset is an unholy mix of blurry anime screen shots and two official art images that's were not entirely 100% in tune with the show. Initial results were rather poor so I had to use generational training to clean up and improve the small dataset. Training a model on its own inputs comes with the risk of "inbreeding" inherent flaws but the result was satisfactory overall. That said this LoRA is not as perfect as I'd like but considering the few models for this character I decided to release it still.General things that may pop up - the PonyXL quality tags will push output towards western style which may cause outputs to deviate from the character design a bit (Hello Tracer). Most commonly emphasizing the "boy" in tomboy too much. Adding the "cute" tag can remedy this. Some parts of her outfit didn't have any good tags, most notably the belt mysteriously floating around her hips can be unreliable. Adding "holster" and/or "pouch" to negatives can help here. The elbow pads can also take different designs depending on what the model decides to draw data from for the specific gen.Naturally, none of this is a much of an issue if you prompt her in different outfits (or none at all).MISCTrained on Pony SDXL at 1024 resolution with a frankenstein'd dataset of 17 images and lots of foul language.All images are untouched txt2img, no inpainting or helper embeddings used for quality transparence, e.g. - things should look even better if you do.The training was done on a small subset of images, as such the output of this LoRA is not an accurate representation neither in quality or style of the source material. Consider buying the official merchandise if you want the good stuff.