I use a 2-card system, so I have a pre-determined single image for every possible combination of two cards in a deck. The values and suits are read phonetically and generate a unique word or phrase for each pair.
So for example, the [7 
][2 
] is “cannon.”
This is one of those pairs that I’ve been able to reduce the subvocalization a lot. All I “say” is the sound “Can” and my brain just kind of sees the rest of the value info and automatically pops up that cannon photo before I “read” any further.
With this technique, a deck can be memorized using 26 images.
This isn’t a PAO, its a single-element system. This means I’m free to craft the interactions however I want between elements and with their locations within whatever memory palace I’m using.
If you’re interested, there’s more about my specific system and how the cards actually translate to sounds / words / images here: