When will we be able to generate manga that consistently maintain character designs, and can generate speech bubbles and fx wherever required?