FireFlow: F ast I nversion of Re ctified Flow for Image Semantic Editing
Yingying Deng, Xiangyu He, Changwang Mei, Peisong Wang, Fan Tang · 2024
Read FireFlow: F ast I nversion of Re ctified Flow for Image Semantic Editing by Yingying Deng, Xiangyu He, Changwang Mei, Peisong Wang, Fan Tang (2024) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2412-07517.txt.
Description
Editing a photo with a text prompt first requires inversion: turning the image back into the noise pattern a generator could have started from. Yingying Deng and co-authors present FireFlow, a method that does this inversion for rectified flow models such as FLUX. Earlier attempts were slow or lost detail, which spoiled the edits that followed. FireFlow needs no retraining and finishes the round trip in eight steps.
The core idea is a carefully designed numerical solver. It matches the accuracy of a costly second-order method while keeping the speed of a simple first-order one. The authors report three times faster running than rival approaches, smaller reconstruction errors, and better editing results. They release the code publicly, aiming to make fast, faithful editing practical for these fast distilled generators.
Catalog record
| CREATOR | Yingying Deng, Xiangyu He, Changwang Mei, Peisong Wang, Fan Tang |
|---|---|
| YEAR | 2024 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 48 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2412-07517