You cannot, but you could let GPT-V analyse the images before hand and then put the description of the image in so it would have the same information