Pretty neat
“This LoRA was trained with colorized manga pages”
Don’t know if you knew, but Pony (and I think SD in general) is actually pretty good at learning from black and white images and applying any colors you want at generation, so straight manga pics are actually perfectly usable in training data. I personally have done a LoRA trained entirely on black and white manga pictures, another LoRA where that was the vast majority of the training data with just a few color pics added, a LoRA where the training data was heavily bulked up with manga, a few more with smatterings of manga, plus a pair of two more where settei lineart was a large chunk of training data. All work just fine as long as you tag it properly. I think it’s rooted in how SD’s learning (likely other models too, but definitely those based on SD) is based on shapes and depth map rather than the color of an image (it can and does learn color, but its its own step).