For consistency, i kept all images with the same phrase, same formatting and same font. and even so it didnt turn out as good as i hoped...

But hey, it does gets it right a good amount of times... and even when it doesnt, its still much more readable

UPDATE: Turns out that, with the same dataset, the Illustrious version managed to do much, MUCH better. The pony one still frequently gets it right, but even so it doesnt compares to the Illustrious version.

So, if possible, you really may want to try the Illustrious version instead. But if not, its no worries, Pony still manages really well.