Nice idea :) I just realized (thanks to your post) that implementing a solution to this is pretty tricky. e.g., the bottom right panel, 1 sentence is split up

Nice idea :) I just realized (thanks to your post) that implementing a solution to this is pretty tricky. e.g., the bottom right panel, 1 sentence is split up into 2 speech bubbles 「この帽子を」＋「お前に預ける」meaning approximately, "I'll leave this hat in your care" Any solution here has to kinda grab the words for a whole panel yet still know how to segment it (and read it RTL, top-to-bottom), and then ideally render something like: "I'll entrust this hat..." + "...to you." or something similar for some sensible split into 2 word bubbles. In the top-left, 「おれの大切な帽子だ」, the translation of "It's my hat" is quite bad. "It's my precious hat." would be much better. Maybe it's an OCR bug? Because LLM should get that nuance pretty easily.