relative_movement: Relative keypoint displacement (Inherit object proporions from the video)
relative_jacobian: Only taken into effect when relative_movement is on, must also be on to avoid heavy deformation of the face (in a freaky way)
adapt_movement_scale: If disabled, will heavily distort the source face to match the driving face
find_best_frame: Find driving frame that best match the source. Split the batch into two halves, with the first half reversed. Gives mixed results. Needs to install face-alignment library.

Part Swap

blend_scale: No idea, keeping at default = 1.0 seems to be fine
use_source_seg: Whether to use the source's segmentation or the target's. May help if some of the target's segmentation regions are missing
hard_edges: Whether to make the edges hard, instead of feathering
use_face_parser: For Seg-based models, may help with cleaning up residual background (should only use 15seg with this). TODO: Additional cleanup face_parser masks. Should definitely be used for FOMM models
viz_alpha: Opacity of the segments in the visualization

Articulate

Doesn't need any

Spline

predict_mode: Can be
- relative: Similar to FOMM's relative_movement and adapt_movement_scale set to True
- standard: Similar to FOMM's adapt_movement_scale set to False
- avd: similar to relative, may yield better but more "jittery/jumpy" result
find_best_frame: Same as FOMM

MRFA

model_name: celebvhq or vox, which is trained on different datasets
use_relative: Whether to use relative mode or not (absolute mode). Absolute mode is similar to FOMM's adapt_movement_scale set to False
relative_movement, relative_jacobian, adapt_movement_scale: Same as FOMM

Installation

Clone the repo to ComfyUI/custom_nodes/

git clone https://github.com/FuouM/ComfyUI-FirstOrderMM.git

Install required dependencies

pip install -r requirements.txt

Optional: Install face-alignment to use the find_best_frame feature:

pip install face-alignment

Models

FOMM and Part Swap

FOMM: vox and vox-adv from

Part Swap

vox-5segments
vox-10segments
vox-15segments
vox-first-order (partswap)

These models can be found in the original repository Motion Supervised co-part Segmentation

Place them in the checkpoints folder. It should look like this:

place_checkpoints_here.txt
vox-adv-cpk.pth.tar
vox-cpk.pth.tar

vox-5segments.pth.tar
vox-10segments.pth.tar
vox-15segments.pth.tar
vox-first-order.pth.tar

For Part Swap, Face-Parsing is also supported (Optional) (especially when using the FOMM or vox-first-order models)

resnet18 resnet18-5c106cde: https://download.pytorch.org/models/resnet18-5c106cde.pth
face_parsing 79999_iter.pth: https://github.com/zllrunning/face-makeup.PyTorch/tree/master/cp

Place them in face_parsing folder:

face_parsing_model.py
...
resnet18-5c106cde.pth
79999_iter.pth

Other

Model Arch	File Path	Source
Articulate	`module_articulate/models/vox256.pth`	Articulated Animation (Pre-trained checkpoints)
Spline	`module_articulate/models/vox.pth.tar`	Thin Plate Spline Motion Model (Pre-trained models)
MRFA (celebvhq)	`module_mrfa/models/celebvhq.pth`	MRFA (Pre-trained checkpoints)
MRFA (vox)	`module_mrfa/models/vox.pth`	MRFA (Pre-trained checkpoints)

Notes:

For Spline, to use find_best_frame, follow above instructions to install face-alignment with its models.

Credits

@InProceedings{Siarohin_2019_NeurIPS,
  author={Siarohin, Aliaksandr and Lathuilière, Stéphane and Tulyakov, Sergey and Ricci, Elisa and Sebe, Nicu},
  title={First Order Motion Model for Image Animation},
  booktitle = {Conference on Neural Information Processing Systems (NeurIPS)},
  month = {December},
  year = {2019}
}

@InProceedings{Siarohin_2019_NeurIPS,
  author={Siarohin, Aliaksandr and Lathuilière, Stéphane and Tulyakov, Sergey and Ricci, Elisa and Sebe, Nicu},
  title={First Order Motion Model for Image Animation},
  booktitle = {Conference on Neural Information Processing Systems (NeurIPS)},
  month = {December},
  year = {2019}
}

@inproceedings{siarohin2021motion,
        author={Siarohin, Aliaksandr and Woodford, Oliver and Ren, Jian and Chai, Menglei and Tulyakov, Sergey},
        title={Motion Representations for Articulated Animation},
        booktitle = {CVPR},
        year = {2021}
}

@inproceedings{
tao2023learning,
title={Learning Motion Refinement for Unsupervised Face Animation},
author={Jiale Tao and Shuhang Gu and Wen Li and Lixin Duan},
booktitle={Thirty-seventh Conference on Neural Information Processing Systems},
year={2023},
url={https://openreview.net/forum?id=m9uHv1Pxq7}
}