The dataset consists of 158 images. The keyword is not required, although it is there.
LoRA is VERY overtrained. Weight 1.0 gives out 3dcg most often according to the type of models from Overwatch. Recommended weights are 0.5-0.7.
I also recommend using a single "detailed" tag, which helps make textures more pleasing. If you need more anime style, then you should add some anime artist to prompt (my personal choice is "asura style")
