I like this layout of text a bit better. Though I think I need to tweak a bit so notes are not so aggressive when not needed.

(Game is Medarots 9)

I think the trick might be to have a DropBox UI element that lets you load different prompts based on how the game is set-up. Like this prompt might be nice for a game with menus and such but if you are just playing a VN like game you would just want dialogue. Could even make prompts for specific games if the text layout is weird.

@quad A tool I am working on that uses a low powered LLM (running locally or on a api) to extract the Japanese text from a image and translate it. The main feature is you can set a directory for it to watch and anytime a new image enters it auto sends it off to the LLM to get processed. So if you are playing something in #RetroArch or another emulator you have this side by side and press the screenshot button when you need a translation. I think I posted a video of a earlier version a week or so ago here:
https://melonbread.dev/notice/AscmWuE6CCbWpvsSZM

Still working some things but will share the repo soon once I work a couple more things out.