Mastering Text-to-Image Diffusion: Recaptioning, Planning, and Generating with Multimodal LLMs