Mask Helper Not showing

[Universal-Invariant]

First, confirm

• I have read the instruction carefully
• I have searched the existing issues
• I have updated the extension to the latest version

What happened?

I have no node Mask Helper. I just installed ReActor and was able to do a faceswap but was looking for the Mask Helper Node and it simply does not show up. Mind you, I am using intel xpu so it is possible that could be an issue(it shouldn't but if there is some cuda check that disables MH then it...).

I have exactly 14 nodes under Reactor:

Fast Face Swap
Fast Face Swap [...
Options
Face Booster
Set Face Swap W...
Save Face Model
Load Face Model
Build Blended Face Model
Make Face Model Batch
Restore Face
Restore Face Advanced
Image Dublicator
Convert RGBA to RGB
Unload ReActor Models

Steps to reproduce the problem

No workflow issue except node doesn't exist

Sysinfo

Windows, xpu, Comfy 0.19.3

Relevant console log

It's not an error, it simply isn't showing.

Additional information

No response

[Buumcode]

I have no node Mask Helper.

Hm. Very strange situation. Try this:

1. Delete your current ComfyUI-ReActor folder: ~\ComfyUI\custom_nodes\ComfyUI-ReActor
2. Take 'freedom' version of Reactor from Codeberg.org: https://codeberg.org/Gourieff/comfyui-reactor-node
   (Top right corner -- Three dots) Download ZIP
3. Unpack ZIP as ~\ComfyUI\custom_nodes\comfyui-reactor-node
   (Folder name is case sensitive)
4. Restart your ComfyUI

Now you should be able to see all this nodes:

***INFO: Reactor Masking Helper protect your hair beauty from 'sharp' effect of Face Restorer (as exampe, after GPEN-BFR-512).

[Universal-Invariant]

Thanks. It is showing now. I cloned the repo and ran install.bat like the original case(out of habit).

Just as an aside: Is there a way to have face stability in a video over frames so that it doesn't distort because it tries to have some continuity between sizes and positions? By doing frame by frame independently if there is a miscalculation it will cause disturbing artifacts.

Also there are no options listed for bbox model. I originally followed the instructions to install yolov5 (I think one was face-yolov5.pt) into the locations set but there seemed to be some discrepancy between the locations. Not sure if it is looking in two places. One was models/ultralytics/bbox and the other models/facedetection. I have the the yolo files(I downloaded higher versions hoping I could use those) in both but still not detected. I also recall that when I originally installed ReActor that it downloaded the yolo models and placed them in the models directory somewhere(one of those directories I mentioned).

[Buumcode]

One was models/ultralytics/bbox and the other models/facedetection

Here is list of models, needed for Reactor:

---C:\ComfyUI\models\facedetection---
detection_mobilenet0.25_Final.pth
detection_Resnet50_Final.pth
parsing_parsenet.pth
yolov5l-face.pth
yolov5n-face.pth
---C:\ComfyUI\models\facerestore_models---
codeformer-v0.1.0.pth
GFPGANv1.3.pth
GFPGANv1.4.pth
GPEN-BFR-512.onnx
GPEN-BFR-1024.onnx
GPEN-BFR-2048.onnx
RestoreFormer_PP.onnx
---C:\ComfyUI\models\insightface---
inswapper_128.onnx
---C:\ComfyUI\models\reswapper---
reswapper_256.onnx
---C:\ComfyUI\models\hyperswap---
hyperswap_1a_256.onnx
hyperswap_1b_256.onnx
hyperswap_1c_256.onnx
---C:\ComfyUI\models\insightface\models\buffalo_l---
1k3d68.onnx
2d106det.onnx
det_10g.onnx
genderage.onnx
w600k_r50.onnx
---C:\ComfyUI\models\sams---
sam_vit_b_01ec64.pth
---C:\ComfyUI\models\ultralytics\bbox---
face_yolov8m.pt

I have the the yolo files(I downloaded higher versions hoping I could use those) in both but still not detected

I've use YOLOv5l for face detection (yolov5l-face.pth).
Face detection is not problem for video face consistency. Face detection works always fine. But after face detectection model for face swapping (i.e. hyperswap_1b_256.onnx) GENERATE a bit different faces every time. And from that, we have 'inconsistency' between frames.

Is there a way to have face stability in a video over frames

No. Reactor make face swap on every frame independently. Which cause disturbing and artifacts. Reactor is IMAGE-based technology. For video face swapping is better to look on some VIDEO-based technology. I don't learn that. May be something help is MoCha

[Universal-Invariant]

I was missing yolov5n-face.pth and GPEN-BFR-1024.onnx and downloaded them and then retried and now a bunch are showing up like yolo26n.pt which I had earlier and did not add. It might have been some caching issue or something strange. I'm using stability matrix to run comfy-ui and it does something funky with models. In any case, it is now showing them a lot more than it was.

I will look into face swap. I wonder though, if it is possible to reduce some of those artifacts? The issue is that the bounding box and T zone is miscalculated in some cases but we know that the in video it is rare that these things chance their positions, sizes, and angles drastically. I'm not sure if there is some way to interpolate the results of what yolo compute but it might help if it is possible. E.g., essentially wrap yolo face detection in something that produces smoother changes to avoid the jump discontinuities that are so obvious and distracting?