By putting it manually do you mean writing it yourself or copying the output of the OCR result? If the latter you can open an issue on the github page and i'll try to have a look at it (also note that you are using a ~free API there).

The final translation is the results of a 3 step process: Detect text position (BOX) -> Extract text from image (OCR) -> translate (TSL).
Of course changing the model itself or its parameters for any of these will yield different results.

Regarding Korean, the only OCR that I've found that supports it is from tesseract which does not give great results. If you have/know-of a better model you can either open an issue on the GH repo so i can add it to the default ones, or add it for yourself only by following the procedure in one of the discussions "Guide on using new models?" on the GH page for ocr_translate.