Stegcracker is designed to crack file made by Steghide. Steghide used the Least Significant Bit (LSB) method. PixelKnot uses an algo called F5 that works by modifying the non-zero DCT values in a JPEG file. Stegcracker will never crack it. My condolences for your electric bill.
I'm scoping out the feasibility to making a similar tool for the F5 algo. The only way we are going to ever crack this thing is by generating and trying millions or billions of passwords. The Java implementation would be way too slow. It's about one second per test. Thus, someone needs to implement a very efficient minimal test in C. Most of it is pretty straight forward. But PixelKnot uses SecureRandom for the all-important psudo-randomness generator. So I need to make a byte-exact recreation of it's behavior. That's where I'm at. The exact code for the SHA1PRNG algo is surprisingly elusive.
It's going to take a while. I am starting to wonder if it's really that important to see whatever trash talk someone was sending to Q.
A slightly shorter route is to just implement it in Java. It woudn't be as fast but we could get it sooner. There is a LOT that doesn't need to be repeated for each test (extracting the image, Huffman decoding, etc). There is room for several orders of magnitude improvement over hitting the Extract function again and again. However, I'd have to learn Java. Some other codefag could probably do this before I finish reading the Java 101 tutorial. Read a line from STDIN, treat it as the password, decode the first few bytes, then if it's the correct sentinel string print the possible password– else continue at top with the next line from SDTIN.