Training-free Diffusion Model Adaptation for V ariable-Sized Text-to-Image Synthesis (Supplementary Materials)