What translation quality can be expected form such "tiny" model, especially for older games with text only in hiragana? Is there any context memory, or is each sentence translated as standalone?
I'll answer to my own question with one fine translation:
"It's called a pine tree, and it's a pine tree. It's a pine tree. It's a pine tree. It's a pine tree."
Hint: original text had no mention of pine trees.
All jokes aside, some sentences were translated correctly and, most important, with only about 1s of delay on CPU.
Nice, I wonder how the banner would work for menu or any complex screen that's not a single sentence.
In the past, I've used MORT to play some Japanese MMO game. In-place overlay makes it easier to actually play the game instead of doing dictionary lookup on the side.
These days I stopped using realtime translator and just embrace the slow pace of learning these languages while enjoying the game story.
Nice! I wrote a script similar to this that triggers when I take a screenshot with Flameshot, OCR's the screenshot using an API and opens web page with the OCR'd text. I prefer to use a Japanese to Japanese dictionary when possible which I have as a browser extension (yomitan). I like that this approach doesn't require an external API.
In the past, I've used MORT to play some Japanese MMO game. In-place overlay makes it easier to actually play the game instead of doing dictionary lookup on the side.
These days I stopped using realtime translator and just embrace the slow pace of learning these languages while enjoying the game story.
MORT: https://github.com/killkimno/MORT
I imagine this is more for turn based games though.
"Security is more important than usability. In a perfect world, no one would be able to use anything!"