FireFlow: Fast Inversion of Rectified Flow for Image Semantic Editing

Yingying Deng, Xiangyu He, Changwang Mei, Peisong Wang, Fan Tang

TL;DR: An 8-step inversion and 8-step editing process works effectively with the FLUX-dev model. (3x speedup with results that are comparable or even superior to baseline methods)

Abstract:

Though Rectified Flows (ReFlows) with distillation offer a promising way for fast sampling, its fast inversion transforms images back to structured noise for recovery and following editing remains unsolved. This paper introduces FireFlow, an embarrassingly simple yet effective zero-shot approach that inherits the startling capacity of ReFlow-based models (such as FLUX) in generation while extending its capabilities to accurate inversion and editing in **8** steps.
We first demonstrate that a carefully designed numerical solver is pivotal for ReFlow inversion, enabling accurate inversion and reconstruction with the precision of a second-order solver while maintaining the practical efficiency of a first-order Euler method. This solver achieves a $3\times$ runtime speedup compared to state-of-the-art ReFlow inversion and editing techniques while delivering smaller reconstruction errors and superior editing results in a training-free mode. The code is available at [this-URL](https://github.com/HolmesShuan/FireFlow-Fast-Inversion-of-Rectified-Flow-for-Image-Semantic-Editing).

Lay Summary:

In the field of artificial intelligence, a technique called Rectified Flows (ReFlows) has been developed to generate images quickly. However, once produced, these images have been difficult to revert to their original form for purposes like editing and recovery.

Our research introduces FireFlow, a method that enhances the capabilities of current ReFlow models. FireFlow allows images to be accurately reversed from their generated form back to their original state in just 8 steps. This is achieved using a specially designed numerical approach, which combines the precision of advanced methods with the speed of simpler techniques.

This advancement has the potential to make image editing and recovery much more efficient and accurate, without requiring additional training. FireFlow is 3x faster than the best existing methods, and it improves the quality of image reconstruction and editing. This represents a significant step forward in making machine learning tools more accessible and practical for real-world applications, helping society to better utilize transformative AI technologies.

Link To Code: https://github.com/HolmesShuan/FireFlow-Fast-Inversion-of-Rectified-Flow-for-Image-Semantic-Editing

Primary Area: Deep Learning->Generative Models and Autoencoders

Keywords: Rectified Flow, Image Editing

Submission Number: 435