the faces are not similar

[ALysandere]

I have been struggling with various errors in the work of ReActor Node for ComfyUI for several days. Thanks for the recommendations on how to eliminate any errors on this resource. Finally, everything worked for me! But it seemed to me that the result should be a little more convincing. Maybe I have not connected something yet? No errors occur.
The following models definitely work: codeformer-v0.1.0.pth, yolov5l-face.pth, retinaface_resnet50.pth, etc. Their influence is noticeable. Maybe the problem is in inswapper_128.onnx? I tried all possible combinations of face indices.

[Gourieff]

The node inputs in your WF are swapped, I think smth is wrong with your WF of with ReActor you use
input_image goes first, source_image - second
And you don't need to crop the face with other nodes

CodeFormer and GFPGAN restoration models are rather bad with blue-eyed people for some reason (they always try to make your char brown-eyed), so if you swap blue-eyed woman to brown-eyed one, try to make your brown-eyed woman (input_image) to be blue-eyed with facedetailer and then do the swap and restore it using GPEN (it can smash details but your char will have blue eyes)
You can apply several restoration steps to gain the result you want

[ALysandere]

Hello! Thank you! I re-uploaded the custom nodes to the workflow and did exactly as you did. But the result did not improve. It seems that instead of the source face, the program substitutes a model face, which has 10-20% influence from the source face. On large workflows using FaceID + Face Detailer after many settings I get a result of about 50% similarity, but so far I can not achieve more. I do not understand why it is not possible to simply rearrange the faces like in Photoshop.

Maybe there is some error in my ComfyUI.
Every time I reopen the workflow with ReActor Node, this element is highlighted as missing. If I click Try Fix in the Manager and reboot, the problem disappears. But the next time it appears again. I have followed all your instructions for updating components for Windows many times (python -m pip install -U pip, -m pip install insightface-0.7.3-cp311-cp311-win_amd64.whl). The only command that does not work for me is venv\Scripts\activate. Even if I run cmd directly from the folder D:\ComfyUI_windows_portable\python_embeded\Lib\site-packages\virtualenv\create\via_global_ref, where venv.py is located, nothing happens (otherwise there is always an error message: "The system cannot find the path specified"

[ALysandere]

This is the message that appears every time ComfyUI is launched.

Cannot import D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-OllamaGemini module for custom nodes: No module named 'google.generativeai'
Traceback (most recent call last):
File "D:\ComfyUI_windows_portable\ComfyUI\nodes.py", line 1931, in load_custom_node
module_spec.loader.exec_module(module)
File "", line 940, in exec_module
File "", line 241, in call_with_frames_removed
File "D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui-reactor-node_init.py", line 23, in
from .nodes import NODE_CLASS_MAPPINGS, NODE_DISPLAY_NAME_MAPPINGS
File "D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui-reactor-node\nodes.py", line 15, in
from insightface.app.common import Face
ModuleNotFoundError: No module named 'insightface.app'

Cannot import D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\comfyui-reactor-node module for custom nodes: No module named 'insightface.app'

[ALysandere]

If insightface.app is a file, then indeed, I don't have such a file anywhere on my computer. And I installed insightface several times, each time it says that all the files are already present

[ALysandere]

Maybe this message will be useful:

Cannot import D:\ComfyUI_windows_portable\ComfyUI\custom_nodes\ComfyUI-OllamaGemini module for custom nodes: No module named 'google.generativeai'
[ReActor] - STATUS - Running v0.5.1-a6 in ComfyUI
Torch version: 2.3.1+cu118

[ALysandere]

I would like something simple and effective, like here. But in this case, it requires many attempts with manual settings for each individual case. And the end result is still not perfect

.