Recent advancements in AI research have introduced significant improvements in text-to-image and text-to-video generation technologies. Researchers from New York University and Facebook AI Research have developed a machine learning method that outperforms traditional ensemble and weight averaging methods by fine-tuning with high dropout rates. Meanwhile, a collaboration between Peking University and Microsoft Corporation has proposed a novel text diffusion model, TREC, that addresses degradation with reinforced conditioning and misalignment issues through time-aware variance scaling. Additionally, the introduction of PixArt-Σ, an advanced Diffusion Transformer-based model, enables the direct generation of 4K resolution images from text through weak-to-strong training. The VideoElevator project has focused on enhancing video generation quality through versatile text-to-image diffusion models. CogView3, leveraging relay diffusion in latent space, offers finer and faster text-to-image generation. Tencent's ELLA model incorporates large language models (LLM) with diffusion models for improved semantic alignment in text-to-image generation. These developments represent a leap forward in the capabilities of generative AI systems, offering more refined, efficient, and semantically aligned outputs.