using a pre-trained LDM sdV.1.5-profund-ckpt, tarin controlnet on Celeb dataset to generate image from image depth, and canny image