ComfyUI-EbSynth

Run EbSynth, Fast Example-based Image Synthesis and Style Transfer, in ComfyUI

https://github.com/jamriska/ebsynth

Powered by FuouM/Ezsynth EbSynth Python Library, which was built from Trentonom0r3/Ezsynth

Workflow

https://github.com/user-attachments/assets/aa3cd191-4eb2-4dc0-8213-2c763f1b3316

ES Translate - ImageSynth

ebsynth_translate.json

ES Video Transfer - EzSynth

You need to set the style frames' absolute indices. The indices should be in ascending order and separated by an underscore (_). Example: 0_1_2_3 for 4 style frames. The order of the style frames will be as how you loaded it. More specifically, it is their index in the batch IMAGES torch.Tensor.

ebsynth_video.json

(This is not a workflow-embeded image)

ES Video Transfer Extra

Three new arguments are added:

1. flow_arch: Architecture of the Optical Flow - "RAFT", "EF_RAFT", "FLOW_DIFF"
2. flow_model: Choose the appropriate model for the architecture.

warp_weight and pos_weight affects the intensity of Optical Flow guides

Read more details and download the models by following the instructions here.

https://github.com/user-attachments/assets/7f43630f-c7c9-40d0-8745-58d1f7c84d4f

Comparison of Optical Flow models

3. source_edge (Optional): You can use edge preprocessors or anything as input. If this is given, edge_method is ignored and will use yours. edge_weight affects this intensity.

ebsynth_extra.json | ebsynth_extra_edges.json

(This is not a workflow-embeded image)

https://github.com/user-attachments/assets/63e50272-aa5c-42a1-a5ec-46178cdf2981

Comparison of Edge methods

Arguments

Refer to FuouM/Ezsynth for full documentation. These are the important ones:

• style_idxes: The absolute indices of the style frames

• edge_method: Edge detection method. Choose from 'PST', 'Classic', or 'PAGE'.

  • PST (Phase Stretch Transform): Good overall structure, but not very detailed.
  • Classic: A good balance between structure and detail.
  • PAGE (Phase and Gradient Estimation): Great detail, great structure, but slow.

• only_mode: Skip blending, only run one pass per sequence Blending essentially run a forward and reverse pass on the same sequence, where the forward uses the first style frame, reverse uses the next style frame.

  Since blending takes a very long time, one may want to just do forward or reverse mode.

  A forward mode is [0----->]. A reverse mode is [<-----0].

  • forward (Will only run forward mode if sequence.mode is blend)
  • reverse (Will only run reverse mode if sequence.mode is blend)
  • Defaults to none.

• do_mask: Whether to apply mask. If False, mask input is ignore. If True, will throw error if mask input is invalid. Defaults to False.

• pre_mask (bool): Whether to mask the inputs and styles before RUN or after. Pre-mask takes ~2x time to run per frame. Could be due to Ebsynth.dll implementation. Defaults to False.

• feather: Feathering for the masked styled frames applying on top of the original frames. Should not be used with pre_mask == True.

• use_gpu_hist_blend: Use Cupy GPU for Histogram Blending (Only affect Blend mode).

  Faster than CPU (~100x). Although the first time may takes longer as it is compiling the kernels. Time taken for both is also negligible. Will be ignored if Cupy is not installed.

  Defaults to False.

Installation

Use ComfyUI-Manager, or manual install.

You don't need to run install.py if you're manually installing it.

git clone --recurse-submodules https://github.com/FuouM/ComfyUI-EbSynth.git

pip install -r requirements.txt

You may also install Cupy and Cupyx to use GPU for some other operations.

Credits

jamriska - https://github.com/jamriska/ebsynth

@misc{Jamriska2018,
  author = {Jamriska, Ondrej},
  title = {Ebsynth: Fast Example-based Image Synthesis and Style Transfer},
  year = {2018},
  publisher = {GitHub},
  journal = {GitHub repository},
  howpublished = {\url{https://github.com/jamriska/ebsynth}},
}

Trentonom0r3 - https://github.com/Trentonom0r3/Ezsynth

https://github.com/princeton-vl/RAFT

RAFT: Recurrent All Pairs Field Transforms for Optical Flow
ECCV 2020
Zachary Teed and Jia Deng

https://github.com/n3slami/Ef-RAFT

@inproceedings{eslami2024rethinking,
  title={Rethinking RAFT for efficient optical flow},
  author={Eslami, Navid and Arefi, Farnoosh and Mansourian, Amir M and Kasaei, Shohreh},
  booktitle={2024 13th Iranian/3rd International Machine Vision and Image Processing Conference (MVIP)},
  pages={1--7},
  year={2024},
  organization={IEEE}
}

https://github.com/LA30/FlowDiffuser

@inproceedings{luo2024flowdiffuser,
  title={FlowDiffuser: Advancing Optical Flow Estimation with Diffusion Models},
  author={Luo, Ao and Li, Xin and Yang, Fan and Liu, Jiangyu and Fan, Haoqiang and Liu, Shuaicheng},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={19167--19176},
  year={2024}
}