[0] https://en.wikipedia.org/wiki/Ge%CA%BDez_script#Ge%CA%BDez_w...
Somewhat oversimplified, but the two Japanese syllabaries hiragana and katakana are around 50 distinct characters each, so that a core 3x4 board of "keys" responding to tap(-and-maybe-swipe (up|down|left|right)) will give you roughly the full set of each. Generally, tapping the key designates the consonant; swiping (or not) gives you the vowel. There are 5 vowels, and roughly 10 consonants. There's a couple of other symbols added on as modifiers for voicing, etc. Again, oversimplified, but that's roughly it.
(Side note, and I'm guessing here, but I suspect this model probably evolved from T9 texting)
From a brief inspection of https://en.wikipedia.org/wiki/Ge%CA%BDez_script, the Ge'ez syllabary/abugida (used e.g. for Amharic) needs 6-8 vowels across at least 26 consonants, and then some more combinations for labialization/velarization, and then some more for application in specific other languages.
Following the Japanese model, that'd be a pretty big grid :) Phonetic input seems a more workable model to me at least.
So probably the user on the video types in a phonetic approximation of the words using the latin alphabet, and the software translates it to the abugida symbols?
Seems plausable, especially because he types several latin characters to get one symbol.
Interesting to note that sometimes he also uses digits.
The library apparently also supports Amharic/Ge'ez, which does use an abugida, but I can't find any videos of this.