Tool RPGM SLR - Offline JP to EN Translation for RPG Maker VX, VX Ace, MV, MZ, and Pictures

[Shisaye]

I've released v2.015.
Fixed a severe error that was causing negative integer code entries to break.
If you for example had a game crash with an error talking about a canvas, even though all you did was edit a choice box, then that was likely the issue.

Also fixed/changed some other stuff, but preventing those crashes should be enough reason to update.
(Check the Changelog attachment all changes.)

 

[ndj865]

For what engine?
For MV it's usually in the plugins.json, for MZ it's usually in the system.json.
For VX, VX Ace it's not standardized in any way, but you can try adding a catchall command to Main of Scripts.

Shisaye
For decrease font size in VX, VX Ace is to put this code:
Font.default_name = "Tahoma"
Font.default_size = 20
You need to open script editor for RPG maker Vx or VxAce and find 'Main' script then add this line to be first line.

Default size of font is 30 and here is View attachment RPGVX_Tutorial_Changing_Font.pdf tutorial for changing Font in RPG Maker VX.

 

[Shisaye]

Shisaye
For decrease font size in VX, VX Ace is to put this code:
Font.default_name = "Tahoma"
Font.default_size = 20
You need to open script editor for RPG maker Vx or VxAce and find 'Main' script then add this line to be first line.

Default size of font is 30 and here is View attachment 5368555 tutorial for changing Font in RPG Maker VX.

That's the catchall method I meant, but you will find that it doesn't work for quite a few titles because they do not use the Font.default_size value in their custom scripts.
In those cases you have no other option, but to manually scan all parts of Scripts and look for their font values in the code.

 

[ndj865]

Shisaye do you know any game with custom script so I can try this method to decrease font size for VX or VXAce game?

Here is the effect of adding the lines "Font.default_name = "Tahoma" Font.default_size = 20 " in the game .
Image one is starting the game without adding above mentioned line
Image two is opening script editor Main script
Image tree adding the line "Font.default_name = "Tahoma" Font.default_size = 20 " in script editor Main script
Image four Effect of line "Font.default_name = "Tahoma" Font.default_size = 20 " in start screen

 

[Shisaye]

Shisaye do you know any game with custom script so I can try this method to decrease font size for VX or VXAce game?

Here is the effect of adding the lines "Font.default_name = "Tahoma" Font.default_size = 20 " in the game .
Image one is starting the game without adding above mentioned line
Image two is opening script editor Main script
Image tree adding the line "Font.default_name = "Tahoma" Font.default_size = 20 " in script editor Main script
Image four Effect of line "Font.default_name = "Tahoma" Font.default_size = 20 " in start screen
View attachment 5368743 View attachment 5368744 View attachment 5368745 View attachment 5368746

A random example would be Missing Link (RJ160066).
That one does not use the Font.default_size value for the text.

 

[ndj865]

I manage to force Font.default_size in game RJ160066
First image is me not touching script editor
Second image is after editing in script editor and font size is 15


I did it by deleting 4 scripts in script editor.
First 共通スクリプトscript that you see when you open script editor, second マルチメッセージ設定 script, third マルチメッセージスクリプト script and first Main script.
Then in remaining Main script add before line rgss_main { SceneManager.run }
Font.default_name = "Tahoma"
Font.default_size = 15

 

[Gwede titit]

For what engine?
For MV it's usually in the plugins.json, for MZ it's usually in the system.json.
For VX, VX Ace it's not standardized in any way, but you can try adding a catchall command to Main of Scripts.

Engine is MV, but i dnt find it. The game is heroin Shirayuki. Helpp me

 

[ndj865]

@Gwede titit

Font size is in plugin.js line 13 for RPG Maker MV.
First image is first line of dialog not edited, fourth image is change of size of the font for first line of dialog.
Fifth image is change of size of the font in red rectangle change size of the font from 28 to 20

 

[Gwede titit]

@Gwede titit
View attachment 5387452 View attachment 5387453 View attachment 5387454 View attachment 5387455 View attachment 5387456
Font size is in plugin.js line 13 for RPG Maker MV.
First image is first line of dialog not edited, fourth image is change of size of the font for first line of dialog.
Fifth image is change of size of the font in red rectangle change size of the font from 28 to 20

Thank you

 

[Nice_II]

I just saw this tool, does it support many engines? Translation++ does not support valkyria games and others i'd like to read.

 

[Shisaye]

I just saw this tool, does it support many engines?

No, SLR only supports RPG Maker, and only VX, VX Ace, MV, and MZ.

 

[Shisaye]

I've released 2.019.
Main change is that I've fixed DSLR error handling for lowercase \c[] \v[] \i[] \n[] and \nw[] commands. (Due to a bug it was only checking uppercase)
If there is an issue it will now always correctly reject the LLM response and on retry give the LLM an exact description of what it screwed up.
Also a whole bunch of other changes since 2.015, but I didn't deem them important enough to make a new post each time.

Relevant Changelog part:
2.019
Fixed DSLR error handling for lowercase commands.
Made broken command reject reasons better understandable to the LLM.
Overhauled wrapper regex patterns to never escape their intended bounds.
Added 3 new global error checks.
If the japanesesource is enabled it will no longer even try to send a full batch or single line to the LLM that does not include any Japanese.
Fixed a word wrapping issue.

2.018
Fixed DSLR \NW[] handling.
I[] commands now get send to DSLR, and are checked for by DSLR.
Changed DSLR system prompts.
Sep error messages are red now.
Sep log messages now show up in the console.
Sep will no longer default to LLM result if it didn't actually include text, while Paddle's result did.

2.017
Changed DSLR escaper entries to work better with individual plugin files.
When using the Fix Cells and Check for Errors button in a SEP project it will now place spaces in empty Intitial Translations cells to prevent drawing of unchanged text.
Changed default max_tokens for SEP to 1200.
Fixed some typos.

2.016
Fixed bug sometimes making negative integers positive.
Made do-duplication ID removal more robust.
Added escaping for do-duplication IDs for the non SLR translator entries.

 

[Shisaye]

I found a really dumb bug with openrouter, and I'm not sure how to fix it, yet.
When you cancel a DSLR batch translation, and then immediately start another one while a response from openrouter was still being generated, it will ignore your new request and instead send you the response for the old request.
In the best case that will crash the batch wasting a bunch of tokens, in the worst case if it had the same line count it will accept the wrong batch and just do rough line by line checks on it. Loosing context, providing a worse or even wrong translation, potentially saving that to cache, and wasting a lot of tokens.

 

[Shisaye]

Another problem is that when a name placeholder like \n[1] is used, and the LLM is told that it's a woman for example.
It will often just replace it with she/her, which is probably fine or even better in a lot of cases, but the system cannot know that.
As a result it will ask for a correction during the line by line checks. Every, single, damn, time. Wasting huge amounts of tokens.

But I have no idea how to handle that better, I could of course just remove the check for name placeholders entirely, but then there's situations in which that would make it accept a cutoff or otherwise broken response. Like it responding with n[1] (without the "\") or something like that.

I also thought about always replacing \n[1] with the actor entry 1, so the placeholder is never sent to the LLM to begin with, but one of the reasons for devs to use those placeholders is so the player can change the character name, and if I remove the placeholders then obviously the name will never change.

 

[boringapocalypse]

When trying to export a translation of images, I just get a line saying "[SEP] Origin path is cache path." and the processing continues indefinitely. I can export other kinds of translated files no issue.

 

[Shisaye]

When trying to export a translation of images, I just get a line saying "[SEP] Origin path is cache path." and the processing continues indefinitely. I can export other kinds of translated files no issue.

Is there more information if you open the console? (F12)
Please send me the trans file and the matching staging data (the folder from www>php>cache). Also the picture(s) you used as input.
Edit: Nvm I found the issue. I had a typo when I made a new log messages and it said "aconsole" instead of console.
Programming is fun.

Edit2: Fixed in v2.021.

 

[Shisaye]

I've released v2.024
I've significantly improved the DSLR retry logic.

Simplified how it works now on default settings:

SLR will send a batch of 10000 characters to the DSLR engine.
DSLR will make smaller batches from it aiming for 1200 characters, but a batch will never ask for less than 40 translations or more than 120 translations.
If the LLM fails to provde a usable response for a batch 3 times it will split the batch in 2 halves and try each 4 times.
If they both succeed it will merge the result back into one batch and do the normal individual translation checks.
Should one of them fail all attempts then it will make individual translation requests for that half, while just running the offline checks on any successful one.

Basically the main change is that it will no longer treat a half batch running out of retries as a failure state and just abort the whole thing, and instead now it actually preserves all successful bits and just does its best to not waste more time/tokens.


Changelog since last release post:
2.024
DSLR will now attempt the second half of the batch even if all attempts for the first half failed.
DSLR will now correctly use all successful batch halfs regardless if the other failed.
Added better console logging for DSLR errors.
Added handling for the response somehow not failing the ok test, but still not being readable for whatever reason. Instead of crashing it will now just retry instead.

2.023
Added D_TEXT processing to DSLR directly instead of the wrapper to better handle fontsize vlaues.
DSLR will now talk about translation requests instead of lines.
Added entry to DSLR split patterns to remove isolation patterns placeholders at the start or end of cells.
Changed default max requests setting to 120.
Pad placeholder setting for DSLR is now turned off by default.

2.022
Fixed it tagging wrongly tagging safe cells red in VX/VXAce.

2.021
Fixed a log message breaking SEP file exporting.
Fixed it sometimes screwing up game titles.

2.020
Improved word wrapping of cells without a closing bracket.
Added support for a new niche script boundary.

 

[Shisaye]

In an episode of sleepy stupid I forgot to actually properly implement the usage of a successful second half of a Full Batch if the first half failed.
Meaning it was claiming to use the second half, but never actually did, or in the worst case used it at the wrong position.
Fixed in v2.026.
If you are using DSLR please update.

Changelog since last post:
2.026
DSLR Fixed it failing to actually use the second half of Full Batch if the first failed.
Fixed the project title and save preset wrongly including the SLRContextID.

2.025
VX/VXAce parser now also collects 'variables' and 'switches' entries.
'switches' get tagged red, 'variables' get tagged yellow.
Added some colors to DSLR progress logging.

 

[Shisaye]

I have some pretty damn good news for someone struggling with running DSLR.
There's a new wave of uncensored qwen instruct models that make models that are actually so impressive at following orders it beats other 32B model variants at 8B, making it run on just about anything.

Qwen3 VL 32B Instruct abliterated v1 I1
for someone with a recent PC (I don't have the hardware to test it a lot, but during my limited testing I have not seen it go into an error loop at all, yet) and
Qwen3 VL 8B Instruct abliterated v2.0 I1
For someone who basically owns "any" CUDA or Vulcan capable GPU.
You can fully offload the Q4_K_S variant on a 1060. (It's seriously only 4.8GB.)

Will the 8B version still occasionally shit the bed and get stuck in an error loop? Yes, but stuff like Gemma3 does that at 27B too, and it failing, and you having to remove some problematic cell from the batch before redoing it, hurts quite a bit less if it's running at like 200 characters per second on a 9 year old GPU.
Because unlike my previous recommendation for offline DSLR this isn't a thinking model. It doesn't waste 5 years each response writing a novel about why something might be translated to "big throbbing cock".
For maximum speed and laziness you could also use the 8B model until it hits an error loop and then just switch to the 32b model for that batch and then go back.
Heck if there's demand I could even automate that, have an automatic fallback model.
I mean since it's offline and free it's not like you would need to worry about token usage.

And as if that wasn't enough, yet, It's a VL model, it has vision. You can use it for SEP picture translations, too. (Although I wouldn't use the 8B variant for that, that probably wouldn't be a meaningful step up from just using paddle.)

I "assume" the thinking model variants would fail slightly less, but I can't imagine the immense speed difference being worth it.
Even premium stuff like GPT or DeepSeek will get stuck in error loops sometimes, you can't really fully avoid that, yet.

 

[Entai2965]

Links for the models mentioned above.

static/unweighted

You must be registered to see the links

You must be registered to see the links

(v1)

You must be registered to see the links

(v2)

You must be registered to see the links

(v2.0)

imatrix/weighted

You must be registered to see the links

You must be registered to see the links

(v1)

You must be registered to see the links

(v2)

You must be registered to see the links

(v2.0)