This model was trained on kohakuXLBeta_beta7, but my device does not have the ability to test XL Lora, so I uploaded the version with the lowest loss value. You can try adjusting the weights to change the rendering effect.