The field of text-to-image models is rapidly advancing, with a focus on improving model controllability and alignment. Researchers are exploring new architectures and training methods to enhance the quality and consistency of generated images. One key area of innovation is the use of structured captions and glyph-conditioned diffusion models, which have been shown to improve text-image alignment and generate more readable and meaningful text in images. Another area of research is the development of new datasets and evaluation metrics for multimodal understanding, including the use of diffusion models as task-aware feature extractors. These advances have the potential to enable more effective and efficient multimodal understanding and generation, with applications in areas such as advertising, education, and creative design. Notable papers include: TextPixs, which introduces a new framework for glyph-conditioned diffusion with character-aware attention and OCR-guided supervision, achieving state-of-the-art results on benchmarks such as MARIO-10M and T2I-CompBench. Vision-Language-Vision Auto-Encoder, which presents a scalable knowledge distillation method from diffusion models, demonstrating cost-efficiency and exceptional performance in captioning tasks.