Tool RPGM DazedMTLTool - A tool that provides quality MTL translations using ChatGPT

[jaden_yuki]

Tencent recently released new translation models: Hunyuan-MT-7B and Hunyuan-MT-Chimera. According to this information -

You must be registered to see the links

- they compete with powerful LLMs and even outperform some.

I haven't looked into it deeply yet, but it seems they are "compatible" with the OpenAI API.

I wanted to test them, but text-generation-webui refuses to load the models using any loaders. I don't have time to deal with it right now, so I might just wait for a GGUF version to check them out.

If the models run fast enough, high-quality machine translation might soon be available to everyone, and we won't have to spend $20-100 on translating each game.

I'm currently using GPT-5 through g4f, with the Copilot provider. I don't know how, but I'm successfully translating even sexually explicit texts. The two models you mentioned are already available at Hugging Face, and they're trending; soon, they'll have inference support. It's useless for translating games for free users, but it can be used to test the models.

You must be registered to see the links

You must be registered to see the links

 

[revyfan]

Tencent recently released new translation models: Hunyuan-MT-7B and Hunyuan-MT-Chimera. According to this information -

You must be registered to see the links

- they compete with powerful LLMs and even outperform some.

I haven't looked into it deeply yet, but it seems they are "compatible" with the OpenAI API.

I wanted to test them, but text-generation-webui refuses to load the models using any loaders. I don't have time to deal with it right now, so I might just wait for a GGUF version to check them out.

If the models run fast enough, high-quality machine translation might soon be available to everyone, and we won't have to spend $20-100 on translating each game.

I did some tests using a Q5 of the non-Chimera version on my 6GB GPU and it's definitely not bad at all Especially for a 7B, and just based off of that size alone is enough to call it above average. Of course, I would still rather use Kimi K2 or Deepseek V3, but if I ever couldn't use them for free, this would be an alternative that can easily fit on my GPU that I wouldn't completely hate.
It still has issues other models have, like translating names wrong and not understanding context, but considering this is 1/70th the size of those other models i've tested, it can be somewhat forgiven.

The picture is the translation it gave me for the explicit Japanese I always give LLMs, not giving it any other direction other than this user prompt: Mimic the 'spoken' (Written) dialect and speech of the untranslated text into English. (If it's 'country', 'girly' 'boyish' 'feminine' ect, translate them into an English equivalent) keep any Japanese honorifics as well and keep to the original format of the text, making sure to translate all non-english text.

This makes me wonder what a 100+B perimeter model using whatever they trained the 7B model would be like, and i'm really hoping to see it sometime soon.

 

[revyfan]

Seem like OpenRouter is giving free access to two new anonymous models (Some have been speculating it's new grok models)
Tried them, and they're fine. Basically Kimi K2/V3 tier for translating. But hey, it's free with over 2M context so it might be pretty good for long form novels/games with very little (If any) censorship/filtering. so use them while you can.

 

[jaden_yuki]

Seem like OpenRouter is giving free access to two new anonymous models (Some have been speculating it's new grok models) View attachment 5226701
Tried them, and they're fine. Basically Kimi K2/V3 tier for translating. But hey, it's free with over 2M context so it might be pretty good for long form novels/games with very little (If any) censorship/filtering. so use them while you can.

but isn't grok 4 level above Kimi K2/V3? Kimi K2/V3 are also free at openrouter. what I'm interested now is a free variation of kimi-k2-0905

 

[revyfan]

but isn't grok 4 level above Kimi K2/V3?

Eh, I was only able to test grok 4 a bit, but yeah it did seem a bit better compared to those models. My personal guess, if these are grok models, is that they're variants/fine tuned versions of Grok 3 that they're going to replace the base Grok 3 with (For both chatting and translating, since they're using Grok 3 for auto translation of tweets now)

 

[Alvillo]

Hello
Excuse me, does anyone know how to use the g4f API with the DazedMTLTool program?
From this GitHub please.

You must be registered to see the links

 

[jaden_yuki]

I'm currently using GPT-5 through g4f, with the Copilot provider. I don't know how, but I'm successfully translating even sexually explicit texts.

the party is over

 

[jaden_yuki]

Hello
Excuse me, does anyone know how to use the g4f API with the DazedMTLTool program?
From this GitHub please.

You must be registered to see the links

the main repo is:

You must be registered to see the links

install it through: pip install --upgrade g4f[all]
run the api through: g4f api
and then an OpenAI-compatible API will be running on localhost:1337

 

[Alvillo]

the main repo is:

You must be registered to see the links

install it through: pip install --upgrade g4f[all]
run the api through: g4f api
and then an OpenAI-compatible API will be running on localhost:1337

Is that all I need to do? And where can I find a compatible API?

 

[jaden_yuki]

what I'm interested now is a free variation of kimi-k2-0905

I tested the kimi-k2-0905, it's even better than the previous version (that was the king of the open source models imo)

 

[revyfan]

I tested the kimi-k2-0905, it's even better than the previous version (that was the king of the open source models imo)

I really only have a few problems with it. Like, sometimes it won't use a sound a character makes (Like Nyaa for 'にゃあ') and it will even ignore commands of me telling it to keep honorifics (Though, admittedly I could fix this by specifcally telling it what Honorifics I DO want it to keep in a translation) but outside of those K2 09 is really good, even better than both V3 version (just barely passing v3.1).

One thing I am starting to notice, however, are LLMs lack of context awareness and hallucinations. Almost all LLMs seem to struggle with using context of a scene for their translations, like how almost all LLMs seemingly get it wrong when translating the Insect Queen story I use (Contributing her brood raping her and not the boy she forcefully mated with) even the ones that do get it right, tend to mess up 30ish% of the time. Don't get me wrong, they are much better than google translate and deepl, but until another huge breakthrough happens, I don't see this issue being fixed anytime soon.

 

[KOX32]

Hi everyone,


I’m having trouble with the tool. I’m trying to translate files from English to Russian, but the tool refuses to recognize the text in my files as translatable, even after I updated LANGREGEX to include English letters.


Every time I try to translate, it just moves the file to the translated folder without translating anything. The log shows:

Input: 0 | Output: 0 | Cost: $0.0000

And debug shows:

2025-09-16 08:04:46,871 - INFO - Weapons.json: Collected 0 names, 0 nicknames, 0 descriptions, 0 profiles, 0 notes, 0 events
2025-09-16 08:04:46,871 - INFO - Weapons.json: Translation complete, tokens: [0, 0]
2025-09-16 08:04:46,871 - INFO - Weapons.json: Translation complete, tokens: [0, 0]
2025-09-16 08:04:46,871 - INFO - Weapons.json: Translation tokens: [0, 0]
2025-09-16 08:04:46,874 - INFO - Wrote translated file: Weapons.json

I'll attach the debug file too.


I’ve tested the API directly, and it works fine, so I guess the problem is with how Dazed reads the files.


Has anyone encountered this issue? How do you prepare files so Dazed actually recognizes text for translation?


Thanks!

 

[jaden_yuki]

grok 4 fast is out! you can use it for free at OpenRouter. it's amazing how this "fast" model can match big open-source models like kimi-k2

 

[thuy2310]

grok 4 fast is out! you can use it for free at OpenRouter. it's amazing how this "fast" model can match big open-source models like kimi-k2

Error code: 404 - {'error': {'message': 'Not Found', 'code': 404}}
Do you know how to fix this? I also use grok 4 fast

 

[MikiyaMori]

I'm currently using GPT-5 through g4f, with the Copilot provider. I don't know how, but I'm successfully translating even sexually explicit texts. The two models you mentioned are already available at Hugging Face, and they're trending; soon, they'll have inference support. It's useless for translating games for free users, but it can be used to test the models.

You must be registered to see the links

You must be registered to see the links

do g4f provide gpt-5 for free?

 

[jaden_yuki]

do g4f provide gpt-5 for free?

it's currently broken, and you'll need some programming experience to fix it. and microsoft's copilot is more efficiently censored now as well. you'll be able to translate safe-for-work texts, but in this case, it's better to use GitHub Models

 

[jaden_yuki]

the new gemini 2.5 flash 09-2025 is better than grok 4 fast

 

[MikiyaMori]

I am having troubles translating unity games, I guess it's just how unity game is made

 

[ChaosXD]

is it possible to make this work with local llms?

 

[dazedanon]

GUI updated to not be shit and good enough for the average user.

Just run START.bat to install everything and run it. If it fails there will be instructions on how to solve it.