Core Concepts
Rectified flow's efficiency stems from retraining with pre-computed noise-sample pairs, enabling its generalization to broader diffusion models as Rectified Diffusion, which achieves superior performance with faster training and lower cost.
Wang, F.-Y., Yang, L., Huang, Z., Wang, M., & Li, H. (2024). Rectified Diffusion: Straightness Is Not Your Need in Rectified Flow. arXiv preprint arXiv:2410.07303.
This paper investigates the key factors contributing to the efficiency of rectified flow in visual generation and proposes a generalized approach called Rectified Diffusion, applicable to a wider range of diffusion models.