Any fellow codefags in the house?
Steganalysis
Since Q has mentioned specifically the PixelKnot steganography tool supported by Google, rather than simply blindly trying stringers and has codes as passwords I have the following suggestion to some Anon who is programmatically-inclined and provide a starting point.
Benfords Law is one avenue of attack against the Faugere F5 algorithm used by PixelKnot, but it appears to require a clean source image for the attack.
J. Fridrich proposed a technique and subsequent updated technique to probabilistically analyze images to detect the presence of steganographic markers:
https://www.researchgate.net/publication/220059082_Breaking_the_F5_Algorith_An_Improved_Approach
A flawed but partial implementation of the original Fridrich technique is located here (Python and c++):
https://github.com/twinz/MscStegano/blob/master/am945-dissertation.pdf
Im proposing taking the dissertation (flawed) implementation and updating it to use the improved algorithm in order to detect the probability of steg messages (and their estimated length) in a given image.
The improved algorithm has true positive and true negative rates of ~80% and ~66% respectively. 80% accuracy is a minimum threshold of accuracy for many machine-learning algorithms and is sufficient for our purposes.
Im proposing running all of the images Q has posted through the updated implementation in order to determine which, if any, contain steg messages, and if so, their length.
Alternatively, there are other open source steganalysis tools available but few if any tailored to the F5 PixelKnot algorithm specifically.