StreamFlow

StreamFlow: Accelerating Rectified Flow

StreamFlow: Theory, Algorithm, and Implementation
for High-Efficiency Rectified Flow Generation

Sen Fang1, Hongbin Zhong2*, Yalin Feng3,
Dimitris N. Metaxas1
1Rutgers University,   2Georgia Institute of Technology,   3Nanyang Technological University
*Equal Contribution
Overview: This video clearly demonstrates the acceleration capability of our Rectified Flow model. It loads a 4-step PeRFlow model and boosts its speed to approximately 600%. Since the remote rendering speed is slower than the generation process, this video is composed of generated frames to match the actual speed.