> In the three examples below, the model accurately renders Japanese, Korean, and Spanish respectively.

And yet the Korean text is not accurate... [1]

[1] E.g. "드레스 컬렉션 dress collection" has vowels ㅔ mixed with ㅐ, "초웜한" should be "초월한 exceeding", "신키한" should be "실키한 silky", "디자언되다" should be "디자인되다 have been designed", "로얼" should be "로열 royal", and so on.