These two ideas – zooming photos, and zooming automata-fields – may be tricky to connect. How do the cellular automata interact with the photos? Is the photo the top level, and each pixel is a full game of life, and there are only two levels? If you run a CA algorithm on a photo, it stops being a photo almost immediately.
If you are interested in nested photos (without CA), you might want to check out this recent project:
For putting CA fields in an image, the main challenge is that, in order to “read” as b/w pixels, your images must either be mostly black or mostly white. So you might want two different display styles of sparse CAs – one with active black pixels on white, one with active white pixels on black.
If you want to cycle back and forth between photos and CAs, you could use the 2001 trick, only bounce – zoom from photo into CA, then zoom from CA out to a new photo, then in, then out. One “trick” used by the 2001 project is to arbitrarily replace the center pixel in the target photo with the next one in the sequence, so that any transition sequence is valid.