ComfyUI-Fluxtapoz
ComfyUI nodes for image editing with Flux, such as RF-Inversion and more
Nodes (22)
Where regional prompting actually enforces its masks
This is the node the whole Fluxtapoz pack is hiding behind
Give one corner of the image its own prompt
The RF-Edit half that records attention before you change anything
Skip the noise round-trip entirely
The RF-Edit half that injects the saved attention and rebuilds
The sampler that skips the inversion and goes straight to the edit
Pick which attention layers the pack's tricks run on
The tiny node that tells Flux how hard to follow your prompt
Turn an image back into noise, semantically
The no-frills inversion sampler for when you just want to unsample
The noise-mixing utility
Walk noise back into your image
Flip your sigmas backwards without tripping over zero
The inverse half of the RF-Edit model wrapper
The outverse half that rebuilds your image after an edit
More detail without a single extra checkpoint
A pass-through node that exists to fight ComfyUI's execution order
Bolt a reference style onto your Flux prompt — with a strength dial
The 19 double-block switches that decide how much of your original survives an RF-Edit
The single-block switches do the heavy lifting
The sharper, fiddlier detail booster
ComfyUI-Fluxtapoz
A set of nodes for editing images using Flux in ComfyUI
Examples
See example_workflows directory for examples.
No ControlNets are used in any of the following examples.
Rectified Flow Inversion (Unsampling from RF Inversion)
Admittedly this has some small differences between the example images in the paper, but it's very close. Will be updating as I find the issue. It's currently my recommended way to unsample an image for editing or style transfer.
Use this workflow for RF-Inversion.
RF-Inversion Stylization
RF-Inversion can also be used to stylize images.
Use this workflow to style images.
RF-Edit (Unsampling from RF-Solver-Edit)
RF-Edit is an alternative way to edit images. It may suit some use cases better than RF-Inversion and I recommend trying both.
Use this workflow for RF-Edit.
Fireflow (Unsampling from Fireflow inversion)
For a faster inversion method there is also Fireflow for image editing.
Use this workflow for Fireflow.
Flow Edit
This is an implementation of image editing from FlowEdit.
It is an inversion free way to edit images.
Use this workflow to get started.
Regional Prompting
Regional prompting allows you to prompt specific areas of the latent to give more control. You can combine it with Redux, but Redux is so powerful it dominates the generation. This implementation is based on InstantX Regional Prompting.
Use this workflow for regional prompting.
Enhancement
There are two nodes for Perturbed Attention Guidance (PAG) and Smoothed Energy Guidance (SEG) that can add detail to images.
The following from left to right: Vanilla Flux, PAG, SEG
Acknowledgements
Thank you to all researchers involved in the tools implemented in this repo.
<details> <summary>Click to see all acknowledgements</summary>@article{rout2024rfinversion,
title={Semantic Image Inversion and Editing using Rectified Stochastic Differential Equations},
author={Litu Rout and Yujia Chen and Nataniel Ruiz and Constantine Caramanis and Sanjay Shakkottai and Wen-Sheng Chu},
journal={arXiv preprint arXiv:2410.10792},
year={2024}
}
@article{wang2024taming,
title={Taming Rectified Flow for Inversion and Editing},
author={Wang, Jiangshan and Pu, Junfu and Qi, Zhongang and Guo, Jiayi and Ma, Yue and Huang, Nisha and Chen, Yuxin and Li, Xiu and Shan, Ying},
journal={arXiv preprint arXiv:2411.04746},
year={2024}
}
@misc{deng2024fireflowfastinversionrectified,
title={FireFlow: Fast Inversion of Rectified Flow for Image Semantic Editing},
author={Yingying Deng and Xiangyu He and Changwang Mei and Peisong Wang and Fan Tang},
year={2024},
eprint={2412.07517},
archivePrefix={arXiv},
primaryClass={cs.CV},
url={https://arxiv.org/abs/2412.07517},
}
@article{kulikov2024flowedit,
title = {FlowEdit: Inversion-Free Text-Based Editing Using Pre-Trained Flow Models},
author = {Kulikov, Vladimir and Kleiner, Matan and Huberman-Spiegelglas, Inbar and Michaeli, Tomer},
journal = {arXiv preprint arXiv:2412.08629},
year = {2024}
}
@article{chen2024training,
title={Training-free Regional Prompting for Diffusion Transformers},
author={Chen, Anthony and Xu, Jianjin and Zheng, Wenzhao and Dai, Gaole and Wang, Yida and Zhang, Renrui and Wang, Haofan and Zhang, Shanghang},
journal={arXiv preprint arXiv:2411.02395},
year={2024}
}