V1
LoRA trained on top-view diamonds, slightly impove vanilla quality. Based on AnythingV3.1 model, good with RealisicVision. Hard overfitted model