Generate top-down view tokens for VTTs. Use "top down view directly from above illustration with white background of a " as prefix and you can add "Leaning forward" if it is not enough top-down.