Interpreter - an AI-powered offline translator for retro Japanese games

screenshot.png

Are there some Japanese exclusive games from days gone by that you always wish you could've played? Did your favourite obscure franchise get a game release in Japan but never saw localization or a fan made translation hack? Well it's your lucky day.

Interpreter is a new project that can live-translate Japanese text in retro games, all offline and without a required internet connection.

The application is optimized for retro games that use traditional pixel-based fonts and uses MeikiOCR. The tool offers two overlay modes - a simple banner overlay as seen above or by replacing text directly over the original Japanese text at OCR-detected positions.

Windows, macOS and Linux installers are available, and after the initial model is downloaded, everything runs offline for privacy and speed.

:arrow: GitHub source
 
I would like to know if you've done such tests by sniffing traffic through WireShark if it connects to one of their servers for, say, collection of user data for training purposes, to atest to their claims of being "offline for privacy". Aside from that, I see this as a pretty useful tool for translating unreleased JRPGs that never made it outside of Japan, only if there's not an English translation of the game I'm playing.
Certified paranoiac millennial moment.
 
  • Like
Reactions: x65943
Certified paranoiac millennial moment.
It's a totally valid concern, especially considering that everything is vibe coded these days, where both the author and users have no idea what the code actually does, only that "it works".

Not like anyone vets open source code anyway, but at least back in the good old days the author knew how the code worked and you could trust it based on their track record.
 
Sabotage or not im not blaming at xlibre dev. The tool offline should make the support for xlibre, because source code its available for everyone to contribute it. If the author of this tool doesnt want to support it, i will respect the dev decision. I will not change to wayland due to perfomance issues.

Per the Github page:

The XLibre contributors strive to clean up and strengthen the existing code base while maintaining backward compatibility to make X11 a viable choice for the future.

If this AI software is X11 compliant (e.g. not utilising a bug in X11) and XLibre is the one not working, that's a bug on XLibre's part and something they should fix, per the whole point of the project.

BTW the software I was talking about doesn't replace X11, instead it implements Wayland within your existing DE for software that needs it. It's how I got Waydroid working on Linux Mint, I think the software is called Weston.
 
It's a totally valid concern, especially considering that everything is vibe coded these days, where both the author and users have no idea what the code actually does, only that "it works".

Not like anyone vets open source code anyway, but at least back in the good old days the author knew how the code worked and you could trust it based on their track record.
Paranoid has been a totally valid concern since the dawn of humankind, and ALWAYS under the SAME circumstances: When you don't understand what's in front of you.

People knowing what they are doing understands. Not knowing what to do means you are the only one to blame for doing something without the proper knowledge.
Your whole comment is a lie, and a really paranoid one. People knows what they develop, and users using it know what it does. Don't blame others for your own lack of knowledge.

This is hosted on GitHub, you can literally see what it does. If you don't know, then don't download it, simple as that.
 
Paranoid has been a totally valid concern since the dawn of humankind, and ALWAYS under the SAME circumstances: When you don't understand what's in front of you.

People knowing what they are doing understands. Not knowing what to do means you are the only one to blame for doing something without the proper knowledge.
Your whole comment is a lie, and a really paranoid one. People knows what they develop, and users using it know what it does. Don't blame others for your own lack of knowledge.

This is hosted on GitHub, you can literally see what it does. If you don't know, then don't download it, simple as that.
How is any part of my comment is a lie? The part about everything being vibe coded? Sure it's hyperbole, but most new thing that gets shared around these days is vibed. Look at anything on HN over the past few months. Look at anything posted in niche coding forums, it's all slop.

Again, no one vets open source code, not even I do (usually). These days I just check if it's vibe coded, if so, then I don't bother using it. It just shows the author doesn't know what they're doing, so I stay far away, while others just remain debuios of it's legitimacy - or paranoid as you call it.
 
How is any part of my comment is a lie? The part about everything being vibe coded? Sure it's hyperbole, but most new thing that gets shared around these days is vibed. Look at anything on HN over the past few months. Look at anything posted in niche coding forums, it's all slop.

Again, no one vets open source code, not even I do (usually). These days I just check if it's vibe coded, if so, then I don't bother using it. It just shows the author doesn't know what they're doing, so I stay far away, while others just remain debuios of it's legitimacy - or paranoid as you call it.
Unless you are supervising literally every single project out there, yes, you are lying. Stretching fake truths you read online doesn't mean you are telling the truth, which you are not.

If you don't check what you use, that's on your bad. Your own little person world is not the real world, kid. Most people out here know what we are doing, we code, we check, we investigate, we test and try. That's why you keep getting new software.
People who understands, checks. People who don't understand, lives in fear.
 
Unless you are supervising literally every single project out there, yes, you are lying. Stretching fake truths you read online doesn't mean you are telling the truth, which you are not.

If you don't check what you use, that's on your bad. Your own little person world is not the real world, kid. Most people out here know what we are doing, we check, we investigate, we test and try. That's why you keep getting new software.
People who understands, checks. People who don't unders
What world are you living in? It's not the real world. I have a hard time believing you have any actual understanding of code and you're just larping. If you do understand, and you enjoy verifying huge amounts of AI slop for every open source project that you use, then more power to you. I don't know a single developer that enjoys this shit, I know that I don't.
 
What world are you living in? It's not the real world. I have a hard time believing you have any actual understanding of code and you're just larping. If you do understand, and you enjoy verifying huge amounts of AI slop for every open source project that you use, then more power to you. I don't know a single developer that enjoys this shit, I know that I don't.
I'm living in the real world, unlike you, apparently. If you think the world has become a huge AI slop, then you are way more paranoid than I thought you were.
 
as my screenshot above shows it still did get things wrong in thomas the thank engine for GBC, but was MUCH better and I actually felt like it added to the experience of the game vs no text
Effectively the application couldn't display three lines at the bottom of the screen when a character (in the case of the screenshot, Edward) is talking, so it ends up being way less accurate than ever.
 
I think the witch Beatrice did it.

On a more serious note meikiocr is actually pretty good, not sure how the results can be this bad lol
it was trained on visual novels so maybe it just doesn’t work well on retro pixelated low-res fonts. Or maybe the translation model is the bad one.
 
  • Like
Reactions: x65943
Not like anyone vets open source code anyway, but at least back in the good old days the author knew how the code worked and you could trust it based on their track record.
It's really easy, actually. The comments are the biggest, easiest way to spot it without diving too deep. AI really likes to comment code like it's explaining everything to a brain dead toddler. I guess it sort of is though.

Call me old fashioned, but I like to know how my program actually works underneath. At least I don't have to deal with surprises and can actually fix the ones that do pop up.
 
  • Like
Reactions: Sir Tortoise
So by that logic if we were to post news about an emulator that claimed 99% compatibility we'd have to test every game ourselves before echoing it?
The solution is quite simple: if you correctly identify a quote as such and also cite the source, no one will think that the statement comes from you.

As far as the AI is concerned, perhaps it just lacks an initial prompt to consider the lines and context together.
 

Site & Scene News

Popular threads in this forum