Please use it with a style lora, it is trained with low quality images, so on its own the results are mid. Works well with any pony style lora. Copy the data from the example images and modify it as you wish