I can't do that either without opening up an image editing tool. Give the model a tool and goal with "vision". Should work better.