One image can produce many perspectives, suitable for training lora or 3D modeling