This particular Lora has vers small dataset of just 24 images (basically almost all just Illustrations from the LN, most of them Black and white as well) which evidently does work, however normally I'd aim for around 100 images at least if possible.

Though what is just as or even more important than the image count is that the images are well cropped and tagged, so there's not much interfering stuff in them.

@tranthai there were enough coloured pictures for that yeah (about 1/4 iirc)