Hence why I mentioned that if you encode the data in the actual image (the part that's guaranteed to survive processing,) you cannot do it with very fine elements, like subtly shifting the colors of individual pixels or the like, because an average Facebook JPEG algorithm, for example, will just destroy that. You need to use data points that could survive heavy JPEG artifacting, and that means very few data points per image, and low bandwidth.

That’s a moving target with no guarantees to stay true. Steganography in the use cases you’ve described adds complexity and additional portability challenges over a plain encrypted file.