Many translators for text in images work one line at a time and then adjust the font size to make the width of the translated text roughly the same as the width of the original text. The variation in font sizes reflects the variation in word lengths across different languages.
I don’t see how this is a valid reason for the problem. In fact that makes it even more nonsensical.
Lines of text are different lengths in different languages. The OCR app scales the it to fit over the original.
Many translators for text in images work one line at a time and then adjust the font size to make the width of the translated text roughly the same as the width of the original text. The variation in font sizes reflects the variation in word lengths across different languages.