I think you misunderstand a little bit how the system is set up. You’re correct that there are 2704 possible pairs of cards (if you allow for duplicate cards.) Shadow cuts the number of images you need to learn to half of that by associating the same image to both a black-first pair and its equivalent red-first pair.
Example:
When you are memorizing the cards, if you see either of these card pairs, the image you need to visualize and memorize is the same. In my case here, both card pairs associate with the image of SaNTa.
In order to differentiate between the two card pairs, you’ll use a method described as a “2-Block” system. Explanation here:
The phonetic structure Lance designed for Shadow will allow you to possibly import 992 of your three-digit images, BUT they need to follow very specific rules. In your 000-099 word set, there can’t be any S-blends to start the words (specifically, no words that start with the SW, ST, SN, SM, SL, STR, SK, or SP sounds.) There also can’t be any words in all of your 000-999 list that start with an H sound because that sound is used in Shadow to designate Kings in the first position.
The normal design of Shadow has nothing to do with visualizing actual shadows. There’s also no rule about using only objects or people or any other type of image. There is no requirement that your images be of any specific type. My original list had a mix of people and objects. There is no rule that designates motion or color or any type of overlay to differentiate red-first or black-first. That is all dependent on position in the image stack at each loci via the 2-block technique.
For more details, check out the following doc. I typed up instructions and tried to clarify all of the different phonetic rules, image stacking, every aspect of the system. There are also tabs in there with worksheets that list every phonetic structure for every card pair and suit combination so that you can fill them in accurately.
Happy to help if you have other questions. Good luck!