metadata
license: mit
pipeline_tag: unconditional-image-generation
MixFlow: Mixed Source Distributions Improve Rectified Flows
This repository contains model checkpoints for MixFlow, introduced in the paper MixFlow: Mixed Source Distributions Improve Rectified Flows.
MixFlow is a simple but effective training strategy for rectified flows that mixes unconditional and conditional source distributions to reduce generative path curvature and improve sampling efficiency. It achieves better generation quality with fewer sampling steps and accelerates training convergence across benchmarks such as CIFAR-10, FFHQ, and AFHQv2.
Resources
- Paper: MixFlow: Mixed Source Distributions Improve Rectified Flows
- GitHub Repository: NazirNayal8/MixFlow
Citation
@inproceedings{
nayal2026mixflow,
title={MixFlow: Mixed Source Distributions Improve Rectified Flows},
author={Nazir Nayal and Christopher Wewer and Jan Eric Lenssen},
booktitle={ICLR 2026 2nd Workshop on Deep Generative Model in Machine Learning: Theory, Principle and Efficacy},
year={2026},
url={https://openreview.net/forum?id=uWktyU3OIJ}
}