A strength of around 1.0 for Lora is recommended.
Description
FAQ
Comments (3)
Uh - is this trained on mosaics / censorship?
YakoooJun 14, 2026
Try adding "censored" and "hidden" to the negative tags. And "prompt," "nudity," "explicit," and the like to the positive tags.
IL excels with hands and anything anatomical. But this LoRA tends to generate constantly 3 finger hands. Results are amazing, and with a bit of Photoshop+inpaint, like anything, it can be fixed. But could you please consider someday checking your dataset or try other/better parameters for the training and post a v2 with less txt2img finger issues?
Thanks nevertheless for all your impressively good trainings ;)