❤️ [2025] "FlowBlending: Stage-Aware Multi-Model Sampling for Fast and High-Fidelity Video Generation" 

[arXived] project page / arxiv

FlowBlending is a stage-aware diffusion sampling method that accelerates video generation by using large models only when capacity matters (early and late stages), and small models elsewhere. It achieves significant speedups while preserving visual quality and temporal consistency.

❤️ [2025] "Syncphony: Synchronized Audio-to-Video Generation with Diffusion Transformers" 

[arXived] project page / arxiv

Syncphony is an audio-to-video generation framework that produces high-quality videos with precise audio-motion synchronization. It improves temporal alignment through motion-aware training and audio-emphasized inference, while maintaining strong visual quality across diverse audio inputs.