Representation Alignment for Generation: Training Diffusion Transformers Is Easier Than You Think