Tool Ren'Py RenLocalizer - Automatic Translation Tool for Ren'Py Games

[konstant61]

I'm also currently unemployed...and I also can't afford many things...

 

[LordGamer00]

I'm also currently unemployed...and I also can't afford many things...

I understand the struggle completely. Since we are in the same boat, I hope you can understand the limitations better.

When I started developing this project, there were no free automatic Ren'Py translation tools available (or the existing ones were very limited). My goal from day one was to create a free tool accessible to everyone. However, financial constraints have followed me my whole life, and I have only managed to bring the project to this point despite these impossibilities.

For the last few updates, I have been trying to fix non-Google engines blindly. I will take one last look at them in the next update. However, after that, I will mark all non-Google engines as 'Experimental'. To be honest, working blindly without the hardware/budget to test them is just wasting time and is mentally exhausting for me.

I wish you the best of luck in finding a new opportunity soon. Let's hope things get better for both of us.

 

[konstant61]

This is the Unicode error that I meant...when I corrected some translation errors in strings.rpy...

 

[LordGamer00]

This is the Unicode error that I meant...when I corrected some translation errors in strings.rpy...

This error is not caused by RenLocalizer itself, but by how you edited the file afterward.

The Problem: When you edited strings.rpy and saved it, your text editor (likely Windows Notepad) saved the file in a different encoding (Windows-1252 or Windows-1251). Ren'Py only accepts UTF-8 encoding, so it crashes when trying to read the file.

The Fix:

Download and install Notepad++ (free)
Open the problematic file: game/tl/russian/strings.rpy
Go to menu: Encoding → Convert to UTF-8 (or "UTF-8-BOM")
Save the file (Ctrl+S)
Try launching the game again.
Pro Tip for the Future: Always use Notepad++ or VS Code when editing .rpy files. Windows Notepad can corrupt non-English characters.

Let me know if it works!

 

[konstant61]

I just wanted to check this...we'll see...

P.S. But first I need to have lunch...)))))))))so to speak, to gain strength...

 

[konstant61]

As far as I know, the standard Unicode encoding is simply UTF-8. I don't know why you installed UTF-8 BOM...

 

[LordGamer00]

As far as I know, the standard Unicode encoding is simply UTF-8. I don't know why you installed UTF-8 BOM...

You are right, standard UTF-8 is usually sufficient and is the industry standard. I only suggested BOM as an alternative because some specific text editors on Windows sometimes handle it better.

However, based on the error log you shared, the file is currently saved as neither. It is saved as Windows-1251 (ANSI). That is why it crashed. So, you can simply save it as Standard UTF-8 (without BOM) as you prefer, and it will work perfectly.

 

[konstant61]

I saved the files in standard UTF-8 mode...without any additions...

P.S. UTF-8 BOM works...Game is running...

 

[LordGamer00]

I saved the files in standard UTF-8 mode...without any additions...

P.S. UTF-8 BOM works...Game is running...

I'm glad to hear that.

 

[denadm]

Hello. I'm facing a dilemma. I use a script bat UnRen or renpy-sdk for language parsing, but in some cases, depending on the quirks of the novel's developer, they don't extract the files to the tl folder. What if I add a separate parser to this program, with an input field for the desired language, but a more in-depth analysis is required. Is that possible? I would be grateful.

 

[LordGamer00]

Hello. I'm facing a dilemma. I use a script bat UnRen or renpy-sdk for language parsing, but in some cases, depending on the quirks of the novel's developer, they don't extract the files to the tl folder. What if I add a separate parser to this program, with an input field for the desired language, but a more in-depth analysis is required. Is that possible? I would be grateful.

I faced the same issue myself: sometimes relying on standard decompilers (like UnRen) or parsing only .rpy files isn't enough.

I discovered that many text strings actually remain trapped inside the compiled .rpyc files and never make it into the decompiled .rpy. Even if you successfully decompile the game, some critical strings (especially dynamic ones in screens or init blocks) are simply missing from the source code. That's why I spent so much time implementing a native RPYC Reader directly into RenLocalizer.

So, yes! Extending the parser is definitely possible and recommended. Here are a few areas you might find helpful based on my experience:

• RPYC & RPYMC Reading: Instead of relying on an external unrpyc tool, writing a direct AST reader (using Python's pickle mechanics) allows you to extract all of the game's data, including those tricky hidden strings. My project does this in `src/core/rpyc_reader.py`. It reads both script bytecode and screen caches (.rpymc).
  
  
• Screen Language (SL2) Parsing: A lot of 'missing' text lives in complex screen definitions (buttons, tooltips, frames). A standard line-by-line parser misses these. You need an approach that understands Ren'Py's Screen Language structures.
  
  
• Technical Filtering: Parsing everything also means getting a lot of garbage code (like file paths or variable names). You'd need a robust filter (I use a symbol density heuristic) to distinguish between 'Dialogue' and 'Code'.

If you want to create a custom parser module, check out `src/core/parser.py` in the repo. It's designed to be modular, so you could easily plug in a 'Deep Analysis' logic there. You can use, adapt, or develop the current RPYC reading logic (`src/core/rpyc_reader.py`) as you see fit. I recommend using it as a basis for your in-depth analyses because writing something from scratch is really headache-inducing. Building on top of something open source is less of a headache. Also, once you're done, I'd love to take a look at your code—if you don't mind, of course. Maybe I'll discover a few tips that I can add to my program.

 

[denadm]

I would gladly use it, but I'm not a programmer. I use AI, but I've used up my two-week limit in the VS Code program, and the alternative AI there isn't as advanced, and I can't afford to pay, given the situation in Ukraine.

It's one thing to pay a couple of bucks for the OpenRouter API and translate a dozen novels, but it's another thing to pay these corporations many times more.

Browser-based Gemini or Copilot can give useful advice, but it's difficult to shuffle code when you don't understand it, and the result is some kind of bullshit.

Therefore, all hope rests on you, if you understand this or have the means to do so.

I noticed an important bug, used deepl. The program does not distinguish between [] and possibly {}.
Example: [RYD] translates as XRPYXVAR0XRPYX and writes an error message: PLACEHOLDER DAMAGED.

Spoiler: Translated log original

You don't have permission to view the spoiler content. Log in or register now.

But when I added {} and [] to the glossary, the translation was successful.

 

[LordGamer00]

I would gladly use it, but I'm not a programmer. I use AI, but I've used up my two-week limit in the VS Code program, and the alternative AI there isn't as advanced, and I can't afford to pay, given the situation in Ukraine.

It's one thing to pay a couple of bucks for the OpenRouter API and translate a dozen novels, but it's another thing to pay these corporations many times more.

Browser-based Gemini or Copilot can give useful advice, but it's difficult to shuffle code when you don't understand it, and the result is some kind of bullshit.

Therefore, all hope rests on you, if you understand this or have the means to do so.

Honestly, I'm not a programmer either. I’d describe myself as a "vibecoder" since I build everything 100% using AI too. I know exactly how painful it is to be stuck with dumb or limited models when the credits run out—it really kills the workflow.

Just a friendly tip: check out "Google Antigravity". It's a new IDE from Google and since it's in preview, it's currently free with much higher limits than standard chat or VS Code trials. It might save you some headache.

Regarding the parser, I'm not sure how much I can improve the filtering right now because I'm currently focused on a more annoying issue specifically with Google Translate. I added an "Aggressive Translation" mode, but it's still not perfect. The tool successfully detects the text and sends it to the API, but for some reason, Google simply returns it untranslated. It’s not a detection failure—the request goes out, but it comes back raw. This happens to about 1-5% of the lines depending on the game.

So, for the next update, I'll likely review the non-Google engines one last time and mark them as "Experimental". Trying to fix or configure them blindly without proper testing is a massive headache.

Spending money on APIs is really tough when the budget is tight. I'm in a similar boat—I managed to snag my current AI subscription at a 1/3 discount, but I have no idea what I'll do once that runs out . This lack of funds really ties my hands; I can't properly test other translation engines because they require paid credits. So, honestly, I don't know if the non-Google engines are working correctly right now. My impression is that it's hit-or-miss; they work for some users and fail for others.

 

[denadm]

I tested Deepl using the free API and OpenRouter using the paid API, given my lack of knowledge about your program's features. But look what I found out:
1. If I parse files and translate them into Translator++ using any source, then when an update comes out, I just translate the new ones.
2. Your program really works great even with Google Translate or any other translation source, but you can't continue, only do a complete re-translation, but it's fast.
I will probably use your program only for some new novels or where there is little text, but since I have already translated about 200 pieces in translator++, I will have to continue using it.

I may test the local AI from the Jan program, but it is unclear when this will occur, given the electricity issues.

 

[LordGamer00]

I tested Deepl using the free API and OpenRouter using the paid API, given my lack of knowledge about your program's features. But look what I found out:
1. If I parse files and translate them into Translator++ using any source, then when an update comes out, I just translate the new ones.
2. Your program really works great even with Google Translate or any other translation source, but you can't continue, only do a complete re-translation, but it's fast.
I will probably use your program only for some new novels or where there is little text, but since I have already translated about 200 pieces in translator++, I will have to continue using it.

I may test the local AI from the Jan program, but it is unclear when this will occur, given the electricity issues.

Hey, thanks a lot for the detailed feedback! Your comparison with Translator++ is spot on, and it actually helped me see exactly what’s missing. Since I'm developing this solo, getting this kind of insight is super valuable.

I’m already working on fixing those gaps. Here’s the plan for the next update:

• Smart Skip: I’m adding a way for the tool to recognize lines you’ve already translated so it’ll only focus on the new stuff. No more full re-translations!
• Resume Feature: I’ll make sure the progress saves as you go. If the power cuts out or something goes wrong, you can just jump back in right where you left off.
• Jan.ai Integration: You can actually use it right now. Just set the Local LLM URL to
  You must be registered to see the links
  in the settings. But to make it easier, I’m adding a dedicated 'one-click' option for Jan in the next version.

Hang in there—I'm pushing these updates soon. Would love to hear what you think once they're out.

 

[csmail12]

Does it already work with the free DeepL API in the latest version? I haven’t tried it yet.

I also tried another model using LM Studio: Deepseek/deepseek-r1-0528-qwen3-8b.
The previous one, qwen3-4b, is way too dumb, so even though the translation runs through, the quality is terrible (but hey, at least it works! ).


So I think for an average PC, the golden middle ground will be qwen-8b. However, after a while it stops translating. The translator feeds it in 50 batches, but after some time it starts throwing 40/50 errors, as if it can’t process one “chunk” of a batch or something like that—I’m not entirely sure.

 

[denadm]

Dude, I'd love to use local AI, but there are three problems:
1. There's a war going on in the country. We're being bombed. The electricity is cut off for 12 hours a day.
2. My laptop can't handle a powerful model, such as gemini-2b.
3. I'm translating into Russian, and not every model can do that; I need accurate prompt.

 

[csmail12]

Dude, I'd love to use local AI, but there are three problems:
1. There's a war going on in the country. We're being bombed. The electricity is cut off for 12 hours a day.
2. My laptop can't handle a powerful model, such as gemini-2b.
3. I'm translating into Russian, and not every model can do that; I need accurate prompt.

I’m very sorry about your situation. I was simply sharing my experience. I thought every bit of feedback counts. For now, I’d be happy even if the DeepL translation worked perfectly.

 

[LordGamer00]

Does it already work with the free DeepL API in the latest version? I haven’t tried it yet.

I also tried another model using LM Studio: Deepseek/deepseek-r1-0528-qwen3-8b.
The previous one, qwen3-4b, is way too dumb, so even though the translation runs through, the quality is terrible (but hey, at least it works! ).


So I think for an average PC, the golden middle ground will be qwen-8b. However, after a while it stops translating. The translator feeds it in 50 batches, but after some time it starts throwing 40/50 errors, as if it can’t process one “chunk” of a batch or something like that—I’m not entirely sure.

Does it already work with the free DeepL API in the latest version? I haven’t tried it yet.

I also tried another model using LM Studio: Deepseek/deepseek-r1-0528-qwen3-8b.
The previous one, qwen3-4b, is way too dumb, so even though the translation runs through, the quality is terrible (but hey, at least it works! ).


So I think for an average PC, the golden middle ground will be qwen-8b. However, after a while it stops translating. The translator feeds it in 50 batches, but after some time it starts throwing 40/50 errors, as if it can’t process one “chunk” of a batch or something like that—I’m not entirely sure.

Glad to hear you're testing out different models! Here is the situation on my end:

Regarding DeepL: I’m about to drop another update for it. It seems to have some trouble preserving tags. Honestly, testing it is a bit of a headache; I tried using the free API for testing, but I hit the character limit before I was even halfway through! By the way, that 500k limit sounds like a lot, but since it's characters and not words, it barely covers a demo game.

Local LLMs & the 'Stop' Issue: You touched on something that’s been bugging me too. I’ve noticed that some models (including APIs) just stop responding at some point, and I haven't figured out why yet. My testing has been a bit limited because my laptop's fans are acting up, and I didn't want to push the hardware too hard. I didn't see any specific errors during my brief tests, but I’ve tried some 'blind' optimizations anyway.

The Hard Truth: To be honest, I can’t guarantee that any engine other than Google Translate will work 100% perfectly, and I’m not sure I can give that promise for the future either. Chasing free API limits to debug things is exhausting to be fair, at this point, rebuilding the entire app sounds easier than dealing with these free-tier restrictions!

So, I don't want to overpromise on the AI/LLM side of things. That said, I’m not giving up; I’ll be 'consulting' with Claude Opus 4.5 and Gemini 3 Pro to see if they can help me spot why the batches are getting stuck. Hopefully, they’ll have some answers!

Dude, I'd love to use local AI, but there are three problems:
1. There's a war going on in the country. We're being bombed. The electricity is cut off for 12 hours a day.
2. My laptop can't handle a powerful model, such as gemini-2b.
3. I'm translating into Russian, and not every model can do that; I need accurate prompt.

I’m really sorry to hear about everything you’re going through. You mentioned your situation before, and I didn't want to press on it too much because it’s such a heavy topic, but I felt the need to reach out. It’s heartbreaking that you have to deal with power cuts and bombings while just trying to work on your projects. No one should have to go through that.

Regarding the program, I wish I could give you a 100% guarantee for the API engines right now. But just like you, I have my own constraints. Since I’m developing this on a tight budget, I have to prioritize my basic needs over buying API credits for extensive testing. I’m basically trying to fix things blindly with whatever free tier I can find.

I really appreciate you sticking with my tool despite the hardware limits and the situation in your country. I’ll keep doing my best to optimize it, even with my limited resources. Stay safe out there—I truly hope things get better for you soon.

 

[csmail12]

Glad to hear you're testing out different models! Here is the situation on my end:

Regarding DeepL: I’m about to drop another update for it. It seems to have some trouble preserving tags. Honestly, testing it is a bit of a headache; I tried using the free API for testing, but I hit the character limit before I was even halfway through! By the way, that 500k limit sounds like a lot, but since it's characters and not words, it barely covers a demo game.

Local LLMs & the 'Stop' Issue: You touched on something that’s been bugging me too. I’ve noticed that some models (including APIs) just stop responding at some point, and I haven't figured out why yet. My testing has been a bit limited because my laptop's fans are acting up, and I didn't want to push the hardware too hard. I didn't see any specific errors during my brief tests, but I’ve tried some 'blind' optimizations anyway.

The Hard Truth: To be honest, I can’t guarantee that any engine other than Google Translate will work 100% perfectly, and I’m not sure I can give that promise for the future either. Chasing free API limits to debug things is exhausting to be fair, at this point, rebuilding the entire app sounds easier than dealing with these free-tier restrictions!

So, I don't want to overpromise on the AI/LLM side of things. That said, I’m not giving up; I’ll be 'consulting' with Claude Opus 4.5 and Gemini 3 Pro to see if they can help me spot why the batches are getting stuck. Hopefully, they’ll have some answers!

Actually, it might be better to focus on improving DeepL first, optimizing it as much as possible. AI is too expensive and not many people use it for that reason. On the other hand, a DeepL subscription isn’t expensive. And if someone wants to get a bit clever, they can create unlimited free accounts.


DeepL is already the middle ground that makes a game enjoyable. It’s fast and not as complicated a system as an LLM.