I know how to photoshop where the background is an image with multiple transparent areas, then bring into vMix for use as top layer in multiview to properly mask out the underlying talent inputs. But, I'd like to do it the CNN way where the background is a video instead of a fixed image. My initial thought is to start with photoshop as in the former above, but instead of image background just make it a color. Then bring it in to vMix as multiview, add in my talent overlays (do you call them "underlays"?), then chromakey out the background color. Finally, start a new input with the desired video and use multiview to overlay the new talent virtual input that was just made in the previous step. Seems like a lot of work. Is there an easier/simpler way to do this?

if you want a video as a top layer with "holes" in it, you need to use a codec that suppors alpha transparency. This can be done with AFter Effects for example and using ProRes as a codec.
This can work but eats a lot of CPU, so I would not advise to do it constantly in a project.

If the video is a short loop, it can be converted to PNG image sequence. Then it will work on GPU hence safely regarding CPU load.

Alternative ways are to go for Luma or Chroma keying but as you noted, it's a lot of work and not very flexible.

It's often easier, safer and quicker to think the other way around, like Doggy suggested. Use the video as a real background and simply add other inputs as layers and adjusting size, position and cropping as needed.

Thanks for the suggestions. Special codecs are way out of my league. I might be able to use some of the built-in Virtual Sets; some but not all allow the background to be user-specified (neither in the VS setup nor in the generated MV). I understand doggy's approach, but you need a very keen eye to exactly line up the sizing/positioning/zooming of the talent within the borders that I intend to surround my talent with. My method in my original post, while I agree cumbersome to some extent, allows me to clumsily add my talent in the holes and quickly size/position/zoom them without fear of impinging on the hole borders. Again, thanks to all for weighing in.

I'm comfortable editing config and other behind-the-scenes files. But, I need to be thinking about others in my organization who are no where as technical as me and will have to maintain the production after I'm gone. I'm confident they will be able to maintain a production that just uses Photoshop (or equiv.) and the vMix UI, but any deeper into vMix than that and I have my doubts. I guess the tradeoff with my original method comes in the extra steps that have now been confirmed by the earlier posts. All that said, I do appreciate hearing from others on how they would approach the question.

The newer versions of vMix have a copy and paste feature for size and position data of layer items. So like others have suggested, make an input that has all your different layer elements in it and then add the video size and position among the other elements. Each element can be copied and pasted so the position, size and cropping data matches exactly for that input and can be used in other layers and inputs. On the layers side tab, choose edit for the layer(right side of layer info), then you'll see the copy and paste icons(hover for the text). This was a game changer for positioning layer elements.