There are SO many things he might have done, with no pre-determined rules. Like, algo-scramble.
Starting with the n-char plaintext, make it a loop. Now move the second letter two places to its right, the third three places, and so on ... until arriving at the original nth letter (painted red?) Or, starting with the digits of pi, move the second letter 3 to the right, the third 1, the fourth 4, und so weiter.
Doing a frequency on 97 weird letters wouldn't help much.
In their example, "HELLO" is the plain text, "XMCKL" is the key, and the ciphertext is "EQNVZ". However, with a one time pad, an equally plausible plain text is "later" with the key "TQURI". Thus, without anymore data, it is simply impossible to know what the original message is.
Was the 10GB video file never released anywhere and is stored in a now bit rotted old HD in your basement?
Reasonable puzzles can be worked out (albeit maybe with a lot of work) with information provided by the puzzle or available somewhere in the environment.
Unreasonable puzzles (like some old Sierra games cough) are impossible without secret inside knowledge by the puzzle maker and/or brute force. And sometimes not even with brute force.
The hash/video example might just be an Easter egg hunt requiring looking across a wide set of videos (somewhat reasonable but boring), or completely unreasonable depending on circumstances.
Edit: Unless the one time pad is a well known relative document, such as the Declaration of Independence.