So those images are pulled directly from the client?
Not exactly.
As you guessed, the final image in the game is obtained by putting 3 or 4 layers on top of each other.
First the background image, which is a simple jpg.
Then the 2 frames.
For upgraded cards, you add the lightening effect in the top-left and bottom-right corner
And finally the specific card image inside the top frame, the text in the bottom frame, the card name and stats.
Unfortunately, I'm not able to reproduce all this offline because I can't find a way to extract the alpha channel from intermediate layers (my software is only able to produce jpg's out of the swf).
So I had to take real screenshots in game and stack them with the background image. Once the background image and the screenshot are exactly on top one another, I just had to remove the written parts of the screenshot and the specific card image and let you see the background by transparency.
P.S. I always feel like i'm totally impossible to understand when I speak technical