What should the content of the generated image be? As written, one might completely ignore the image content and generate a blank image of the required size.

And what exactly counts as metadata stored in the image that is used to help the program? If the image format explicitly contains the image size (most image formats do), does that already count as forbidden metadata? If not, I can't imagine what information would be helpful.

On the other hand, it may be impossible to get the image size without reading that explicit size (as e.g. an 80x25 image has exactly the same number of pixels as a 40x50 image, and the explicit numbers may be the only information telling how to split the pixel data into rows).