Self-Corrected Flow Distillation for Consistent One-Step and Few-Step Text-to-Image Generation

TL;DR

This paper introduces a self-corrected flow distillation method that enhances one-step and few-step text-to-image generation, achieving consistent high-quality results efficiently.

Contribution

It pioneers the integration of consistency models and adversarial training within flow matching for unified, efficient, and high-quality generative modeling.

Findings

Achieves consistent generation in both one-step and few-step sampling.

Outperforms existing methods on CelebA-HQ and COCO benchmarks.

Reduces the number of function evaluations in sampling.

Abstract

Flow matching has emerged as a promising framework for training generative models, demonstrating impressive empirical performance while offering relative ease of training compared to diffusion-based models. However, this method still requires numerous function evaluations in the sampling process. To address these limitations, we introduce a self-corrected flow distillation method that effectively integrates consistency models and adversarial training within the flow-matching framework. This work is a pioneer in achieving consistent generation quality in both few-step and one-step sampling. Our extensive experiments validate the effectiveness of our method, yielding superior results both quantitatively and qualitatively on CelebA-HQ and zero-shot benchmarks on the COCO dataset. Our implementation is released at https://github.com/hao-pt/SCFlow.git.