Enhance old or low-quality images in ComfyUI. Optional features include automatic scratch removal and face enhancement. Requires installing a number of small checkpoints and VAEs.
Based on microsoft/Bringing-Old-Photos-Back-to-Life.
Before installing, make sure that any virtual environment is activated and if needed, prepend the target python executable before the pip command.
cd ./ComfyUI-Bringing-Old-Photos-Back-to-Life/
path\to\python.exe pip install -r requirements.txt --upgradePyTorch and TorchVision are also required, but should already be installed. (See the main ComfyUI repo for installation details.)
You may encounter issues while trying to install dlib on Windows. Here are some possible solutions:
- Manually install requirements.
You may need to have CMake already installed on your system and CMake\bin added to your PATH to build Dlib.
path\to\python.exe -m pip install easydict matplotlib opencv-python scikit-image scipy
path\to\python.exe -m pip install cmake
path\to\python.exe -m pip install dlib==19.24.1- Find pre-built wheels online.
Alternatively, some people have pre-built wheels for Dlib, however, there may not be any built with your version of python.
path\to\python.exe -m pip install path\to\dlib.whl
Set device_ids as a comma separated list of device ids (i.e. 0 or 1,2). Use -1 for cpu.
Place in models/vae/.
- restoration/VAE_A_quality/latest_net_G.pth
Extract the following models and place them inside models/vae/.
- restoration/VAE_B_quality/latest_net_G.pth
- restoration/VAE_B_scratch/latest_net_G.pth (scratch_detection)
Extract the following models and place them inside models/checkpoints/.
- restoration/mapping_quality/latest_net_mapping_net.pth
- restoration/mapping_scratch/latest_net_mapping_net.pth (scratch_detection)
- restoration/mapping_Patch_Attention/latest_net_mapping_net.pth (mapping_patch_attention)
Extract the following models and place them inside models/checkpoints/.
- detection/FT_Epoch_latest.pt
Download - shape_predictor_68_face_landmarks.dat
Extract the following models and place them inside models/facedetection/ (custom directory).
- shape_predictor_68_face_landmarks.dat
Download - Face Enhancement Models
Extract the following models and place them inside models/checkpoints/.
Set device_ids as a comma separated list of device ids (i.e. 0 or 1,2). Use -1 for cpu.
- Setting_9_epoch_100/latest_net_G.pth (256x256)
- FaceSR_512/latest_net_G.pth (512x512)
Using ComfyUI-Image-Round Nodes
Running the models may fail if image dimensions are not multiples of 8 or 16 pixels (depends on the model used).
- Images may need to be scaled/cropped/padded to the nearest 8 or 16 pixels to avoid a crash. (Use something like my other ComfyUI-Image-Round nodes.)
- "Detect Faces (Dlib)" and "Enhance Faces" nodes will currently return the original image if no faces were found.
@inproceedings{wan2020bringing,
title={Bringing Old Photos Back to Life},
author={Wan, Ziyu and Zhang, Bo and Chen, Dongdong and Zhang, Pan and Chen, Dong and Liao, Jing and Wen, Fang},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
pages={2747--2757},
year={2020}
}@article{wan2020old,
title={Old Photo Restoration via Deep Latent Space Translation},
author={Wan, Ziyu and Zhang, Bo and Chen, Dongdong and Zhang, Pan and Chen, Dong and Liao, Jing and Wen, Fang},
journal={arXiv preprint arXiv:2009.07047},
year={2020}
}The codes and the pretrained model in this repository are under the MIT license as specified by the LICENSE file. We use our labeled dataset to train the scratch detection model.
This project has adopted the Microsoft Open Source Code of Conduct. For more information see the Code of Conduct FAQ or contact opencode@microsoft.com with any additional questions or comments.