Then just use the text file as a "counter nonce" and count from 0 until we find a counter that hashes to the vanity prefix we want.
The advantage of this is flipping the capitalization of the first letter of each word doesn't drastically alter the visual appearance of the file (tho it may alter semantics!), and you get an easy to understand amount of entropy (number of states) to walk through.
The C version linked only supports counting up to ~4 billion but it could be easily extended. A JS version using BigInt can go unlimited.
For reference the included example text took a surprising 2 billion iterations to discover the correct 8 digit hash which is about what you'd expect from random (i.e., half the state space of 16*8 = (2*4)*8)