Shotcut Version 22.01.30 and previous version tested.
It seems that the Reduce Noise: Quanization filter ignores:
1/ masks filters and
2/ chromakey filters
1/If Reduce Noise: Quanization is placed after a mask the Reduce Noise: Quanization filter will still be applied to the whole image area.
2/ If the Reduce Noise: Quanization is placed after a chroma key filter, no matter what other filters are also applied before the Reduce Noise: Quanization filter the filter will mess with the chroma key filter. It will either show the original green screen or will just show a black screen instead of your background.
Thanks, and there is an easy fix I can include for the next version 22.03.
You can fix it in your install (sorry, not possible for macOS or Linux AppImage, Snap, and Flatpak users) by editing the file Shotcut\share\mlt\avformat\yuv_only.txt and add a line with only fspp. Save and restart Shotcut if it is running.
It did not seem to affect the mask issue, but I really should have put that on a different report.
As it turns out, I ended up using the quantization filter on the entire screen area, so it makes no difference for this current usage for me.
I tested it using Mask: Simple Shape. The bug was the filter reseting the alpha channel to fully opaque, which is most observable using Alpha Channel: View. That problem affects anything creating an alpha channel such as a mask or chroma key. Did not even bother to test chroma key.
I do see an issue when I try again what I was attempting, but it may just be me.
I had 2 Mask: Simple Shape filters. 1st overwrite and the second add. If I do this it seems to not do what I was expecting. Is it not possible to add 2 masks?
In the 2 mask image I sent, for some reason when the second mask is there it then just shows the 2 mask areas (rect and ellipse) instead of the actual full person like in the very first image I sent. Maybe it’s me, but I could not see a way to have the 2 masks and still display the full person.
I don’t really need this now as mentioned, I have applied the quantization to the full image anyway now.