UniEdit-Flow
Unleashing Inversion and Editing in the Era of Flow Models

Guanlong Jiao^1,3, Biqing Huang¹, Kuan-Chieh Wang², Renjie Liao³

¹Tsinghua University, ²Snap Inc., ³The University of British Columbia

TL;DR: A highly accurate and efficient, model-agnostic, training and tuning-free sampling strategy for inversion and editing tasks. Support text-driven image 🎨 (FLUX, Stable Diffusion 3, Stable Diffusion XL, etc.) and video 🎥 (Wan, flow-based video generation model) editing.

💜 Overview

In this work, we introduce a predictor-corrector-based framework for inversion and editing in flow models. First, we propose Uni-Inv, an effective inversion method designed for accurate reconstruction. Building on this, we extend the concept of delayed injection to flow models and introduce Uni-Edit, a region-aware, robust image editing approach. Our methodology is tuning-free, model-agnostic, efficient, and effective, enabling diverse edits while ensuring strong preservation of edit-irrelevant regions.

✨ Feature: Text-driven Image / Video Editing

More results can be found in our project page.

🎨 Image Editing

Editing Prompt	Source Image	FLUX	Stable Diffusion 3	Stable Diffusion XL
A long short haired cat with blue eyes looking up at something.
Two origami birds sitting on a branch.
A clown in pixel art style with colorful hair.

🎥 Video Editing

Editing Prompt	Source Video	Wan + Uni-Edit
A young rider wearing full protective gear, including a black helmet and motocross-style outfit, is navigating a BMX bike motorcycle over a series of sandy dirt bumps on a track enclosed by a fence...
A koala cat with thick gray fur is captured mid-motion as it reaches out with its front paws to climb or move between tree branches, surrounded by lush green leaves and dappled sunlight in a forested area.

👨‍💻 Implementation

Here we provide two implementation options:

• Implementation by diffusers: Support FLUX (e.g., black-forest-labs/FLUX.1-dev), Stable Diffusion 3 (e.g., stabilityai/stable-diffusion-3-medium), Stable Diffusion XL (e.g., SG161222/RealVisXL_V4.0), etc., for text-driven image editing tasks. As well, support Wan (e.g., Wan-AI/Wan2.1-T2V-1.3B-Diffusers) for text-driven video editing tasks.
• Implementation on official FLUX repository: Implementation based on original FLUX. The performance is slightly better than the diffusers-based FLUX pipeline.

🔮 Acknowledgements

We sincerely thank FireFlow, RF-Solver, and FLUX for their awesome work! Additionally, we would also like to thank PnpInversion for providing comprehensive baseline survey and implementations, as well as their great benchmark.

📑 Cite Us

If you like our work, you can cite our paper through the bibtex below. Thank for your attention!

@misc{jiao2025unieditflowunleashinginversionediting,
    title={UniEdit-Flow: Unleashing Inversion and Editing in the Era of Flow Models}, 
    author={Guanlong Jiao and Biqing Huang and Kuan-Chieh Wang and Renjie Liao},
    year={2025},
    eprint={2504.13109},
    archivePrefix={arXiv},
    primaryClass={cs.CV},
    url={https://arxiv.org/abs/2504.13109}, 
}