Hi Clare -
Tom, do you do this only with adjustment layers as opposed to putting two masks onto a layer with a pixel-based image? If so, why would you do this?
No, the technique I described above also works with ordinary pixel layers, not just adjustment layers.
I would use two masks on a pixel layer for exactly the same reason as one might do it on an adjustment layer, ie, you want a two step selection process where the first selection / mask is refined further by the second selection/mask.
I can see maybe doing it by copying a layer, then merging them. (though I always group original adjustments of any kind and turn them off; I hate to not be able to go back and make changes; or make a smart obj.)
I try to use adjustment layers (not pixel layers) as much as possible for a number of reasons including future editability (as you correctly suggest), and doing so reduces memory requirements.
And if you should add a 2nd mask to the pixel layer, how do you do that? I always end up with a vector mask when I add an additional. Would the mask be done similarly? On a second layer above?
As you describe, one always gets a vector mask when one tries to add a 2nd mask to a layer. That's why I do it as I described above, ie, putting the 2nd pixel-based mask on a folder that includes the layers with their own pixel-based masks.
When I'm working on a completely photographic image, I rarely end up using any pure vector selection masks or related techniques. I might start a selection with an outline made with the pen, but then I usually rasterize it so I can use techniques such as "refine edges" and similar techniques on the mask to keep a nice organic feel. In contrast, if I'm doing vector art, it's the opposite - I rarely use pixel based selections. However, if I'm doing vector art, I'm usually working in Illustrator, not Photoshop.
But can you even add a mask to a blank layer.
You can. Actually, it is a common technique is to put the needed mask on an adjustment layer (eg, levels) set to do nothing. This saves lots of memory compared to masking an intermediate pixel-based layer.
However, the technique I favor is the mask on a folder (as described in my earlier post). Even though you can get similar memory savings from both techniques, the reason I favor this over a mask on a do-nothing adjustment layer is because it saves computational steps, and hence processing time. My understanding is that all the same mathematical computations must be done (ie, with zero'ed parameters) even if an adjustment layer is set to do nothing. However there are no mathematical computations involved if a folder is placed around a single or group of layers. The folder merely tells the program how to group the operations and in what order to execute them, but doesn't introduce any extra math.
To complete the answer to your question about using a mask on a blank layer, of course you can do this, but it's usually not very useful unless it's not really a truly blank layer and/or the blend mode set to something other than "normal". An example of this might be a masked solid color layer set to "color" blend mode.
HTH,
Tom