A LoRA trained on a selection of (subjectively) the best images trending on DeviantArt. It achieves quite saturated / dramatic / cinematic images and works best on top of an already good performing checkpoint.