generated at
ReActor Node for ComfyUI
2025/1/17
>This repository has been disabled.

インストール
ComfyUI Managerを使ってインストールするだけではインストールできない
ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui-reactor-node にある install.bat を実行
一度ComfyUIを再起動
多分失敗すると思うので → ComfyUIにInsightFaceをインストールする

ver0.5.2
ReSwapperモデルに対応

顔復元としてCodeFormer/GFPGAN/GPEN-BFRが使用可

基本的には顔復元に関するパラメータを別のノードにまとめただけのもの
ただし restore_with_main_after が追加される
ReActorの中の人のinsight faceは128px × 128pxの画像しか出力できないため、それを元画像のサイズに合うように拡大してペーストする
顔復元モデルを使う場合は、"画像を拡大した後"に処理が行われる
restore_with_main_after true にすると"顔復元モデルで処理された後"に画像の拡大をする
こちらのほうが滑らかな画像になる
とされているけれど、五十歩百歩nomadoor

複数枚の参照画像を元にFaceモデルを作り、それでFace Swapできます
すべて同じ人物にすれば精度が高くなる(かもしれない)
別人の画像を組み合わせれば新しい顔が作り出せる


error
EP Error D:\a\_work\1\s\onnxruntime\python\onnxruntime_pybind_state.cc:743 onnxruntime::python::CreateExecutionProviderInstance CUDA_PATH is set but CUDA wasn't able to be loaded. Please install the correct version of CUDA and cuDNN as mentioned in the GPU requirements page (https://onnxruntime.ai/docs/execution-providers/CUDA-ExecutionProvider.html#requirements), make sure they're in the PATH, and that your GPU is supported.
こんな感じのエラーがでたとき pip install -U onnxruntime onnxruntime-gpu したら治ったnomadoor
portable版のコード
$ cd path\to\ComfyUI_windows_portable
$ .\python_embeded\python.exe -m pip install -U onnxruntime onnxruntime-gpu