The Mystic Dragoons (J+Partial English) Beta 1.1 (Resumption of an abandoned project )

[1vierock]

*Update *

 

[awaite85]

Lot of people will appreciate this!

 

[Escaeva]

Absolute gratitude and full support! Thank you, it looks amazing!

 

[marxcore]

Keep up the great progress! :)

 

[FillerBunnyNinjashark]

We all know what it takes, which is translate the script. And that's what this project isn't doing if it's using AI. The rest is romhacking. Unfortunately, there's always been more romhackers then translators.
I have no ill will against the guy, personally. I'm just disappointed. If he wants to do a AI translation, he can, although I believe projects like this should always be labelled as AI so people don't waste their time. A AI translation from Japanese does not work well enough to result in a good project, so I don't really care if it gets dropped either.
"No you have to take Japanese lessons for 5 years then move there so you can really learn it before you can say AI translations are bad even though it's obvious from the English results!!!" is not a good argument.

So he needs an editor then

 

[UnDubious]

So he needs an editor then

A translator. If you start from a machine translation there are errors that an editor won't be able to catch.

 

[mighthyman]

Ai translations are extremely good today, if you use a state of the art model. A bunch of "official" translator were caught using some to do their work.
might as well just skip the middle men

 

[UnDubious]

Ai translations are extremely good today, if you use a state of the art model. A bunch of "official" translator were caught using some to do their work.
might as well just skip the middle men

Where are these state-of-the-art models for Japanese to English? Every auto translation I've seen is horrible when it comes to translating dialogue and narrative text from Japanese. And which translators got caught using AI for Japanese to English? Can you link to some stories? And did they get caught because the translations were bad in the first place?

 

[TruxtonTatsujin TruxtonTatsujin]

Ai translations are extremely good today, if you use a state of the art model. A bunch of "official" translator were caught using some to do their work.
might as well just skip the middle men

by that logic why don't we all just use google lens to play everything and skip the OTHER middle man.

 

[mighthyman]

Where are these state-of-the-art models for Japanese to English? Every auto translation I've seen is horrible when it comes to translating dialogue and narrative text from Japanese. And which translators got caught using AI for Japanese to English? Can you link to some stories? And did they get caught because the translations were bad in the first place?

Sure the visual novel community has been revaluating ai models for a while now, and constantly is impressed: https://www.reddit.com/r/visualnove...difference_14_years_makes_in_offline_machine/ .

The official translators only got caught because they left the AI prompt in. AI translation's are soo good now even most official translators are simply doing light editing. You need a good translation prompt and need to use a state of the art model like gpt 5, claude, gemini 2.5, and grok and you also need to make sure they are in high thinking mode. Google gemini 2.5 nano banana is almost at the place where it can even translate manga pages and KEEP THE FONT STYLE in the translated image, but you still need to code avoid some issues.

Ignore the haters, they are slobbering retards. Try it for your self at be amazed, but you HAVE to pay money for premium models. The IQ difference between the state of the art models and the "free" models is HUGE.

Post automatically merged: Oct 11, 2025

by that logic why don't we all just use google lens to play everything and skip the OTHER middle man.

Maybe in the next 5 years. Google Lens kinda sucks right now, Retroarch has that feature or one like it built in already, if you create a custom server to deal with minor issues. It seems like the only reasonable way to play shin megami tensei 2 and if for the psx right now. Tons of youtube videos on how to do all of that.

i prefer to hack into the actual systems so i can match the font for now. But in spaghetti code games it might be your only option for now.

 

[Remisse]

Sure the visual novel community has been revaluating ai models for a while now, and constantly is impressed: https://www.reddit.com/r/visualnove...difference_14_years_makes_in_offline_machine/ .

The official translators only got caught because they left the AI prompt in. AI translation's are soo good now even most official translators are simply doing light editing. You need a good translation prompt and need to use a state of the art model like gpt 5, claude, gemini 2.5, and grok and you also need to make sure they are in high thinking mode. Google gemini 2.5 nano banana is almost at the place where it can even translate manga pages and KEEP THE FONT STYLE in the translated image, but you still need to code avoid some issues.

Ignore the haters, they are slobbering retards. Try it for your self at be amazed, but you HAVE to pay money for premium models. The IQ difference between the state of the art models and the "free" models is HUGE.

The guy working on this project is getting shit on not because he's using LLMs to translate the game, but because he's doing it without the scrutiny of actual translators who would be able to catch most mistakes. Even the person who wrote the post you linked is saying people should learn Japanese if they can so they can spot those errors.

You keep claiming that official translators only do "light" editing nowadays, yet you won't provide any actual sources showing who these official translators are or how "light" this editing process truly is? And no, Reddit posts from fan translators are not exactly what I'd define "official", unless you're using the word improperly and meant something else entirely.

You can bitch and throw r-slurs around all you want, it still won't change the fact that even state-of-the-art models cannot be blindly trusted by people who don't know better. Yet from the way you're responding, I can only assume you want to defend this guy's work on principle alone and certainly not because he's doing a good job (he's not). But don't let me stop you from enjoying some quality slop.

 

[UnDubious]

The translator who got caught was doing English to Spanish. I knew it would either be this or just a garbage translation in the first place. Again, translation between Western languages is way easier and comes out better. It's nothing like Japanese. Then there's a video showing how bad it was years ago, and showing that it's still pretty bad today. Reading AI dialogue from Japanese today, even "state-of-the-art", if that's what this is, feels like every character is a weird idiot with mental issues. It doesn't properly convey the characters at all. It just gives you a good idea what's being said, most of the time. That's not good enough for a RPG or any game where the story matters. And then you have the pronoun errors, since they're often omitted and have to be inferred based on context in Japanese. The machines still attribute them at random, or even misattribute actions to subjects so you come out with garbage.
You can plainly see in the screenshots for Mystic Dragoons that a lot of it just reads wrong.

 

[TruxtonTatsujin TruxtonTatsujin]

Generative AI is one of the worst things to ever happen to modern culture. There are far to many people who want to blindly trust it. They use it because they're too lazy to do the work themselves and think the AI's output is "good enough". It's not, and the fact that so few people have an eye for quality scares me.

 

[mighthyman]

This ain't college and I'm not your professor, you can use google to find sources, but I'll do it this once for all the people who just talk/complain a lot and don't actually do the hard work like 1vierock is.
Anyone calling AI as slop makes them instantly irrelevant.

Here's the links people couldn't google for themselves:
- Crunchyroll Accidentally Reveals They've Been Using ChatGPT for Sub Translations (apparently without even light editing) https://www.cbr.com/crunchyroll-chatgpt-sub-anime/
- https://www.reddit.com/r/visualnovels/comments/1heuvkf/the_difference_14_years_makes_in_offline_machine/ ( the reddit link i already posted)
- https://www.reddit.com/r/visualnove...ude_3_sonnet_translation_sample_of_an_entire/ (note from reddit: I've tested quite a few LLMs over the last months and I am quite impressed at what Claude 3 Sonnet can do. It seems to get the context right 99% of the time, unlike ChatGPT4 which trips up quite often.)
- https://www.reddit.com/r/visualnovels/comments/1itjthy/why_are_ml_translations_still_looked_down_upon_in/ (in 2025. . . on par with / sometimes surpass fan translations” when used with context; good for hobbyist perspective.)
- https://www.japantimes.co.jp/life/2023/07/18/language/chatgpt-bing-bard-and-deepl-which-one-offers-the-best-japanese-to-english-translation/ ( this is old and already shows how far we've come. LLMs for translation blow the old MT techniques like deepl away)
- https://news.ycombinator.com/item?id=40219765 (I find ChatGPT (4) to have excellent Japanese skills already. It seems to produce more accurate and certainly more idiomatic translations than Google Translate does, and can even explain its own translations afterwards.)
- https://www.reddit.com/r/JuJutsuKai...is_a_mistranslation_the_official_translation/ ( one user said a gpt 4 translation made more sense that the official)

The worst the AI models will be is today. State of the art models like gpt-5, gemini 2.5 just keeping getting better and better.If you don't like reddit, and your asking for official sources on a field that is rapidly evolving, that is a you problem. Most scientific articles on the process are years behind still publishing on gpt 3, just because of the scientific publishing process is time consuming.

You can always pay 100 bucks and try the AI's yourself.

Finally if you don't like the work, go start your own team to translate games. Hacking into old games is insanely hard especially psx games. You will never be able to do a great translation on psx games because you have to worry about byte matching (unless you are willing to write your own custom mips overlays).

Its comments like on this discussion which makes me realize why the Shin Megami Tensei PSX hacker who translated the game just left the community. Which is an UNFORTUNATE THING.

 

[kioskmanagerBERT]

Y’all complaining about the translation like have you’ve actually played old PS1 games. Those were straight-up garbage translations half the time.
imo people are just freaking out because they saw “machine translation” and lost their minds.
I finished 1vierock’s partial english Panzer Bandit translation and it was totally fine.
This one’s gonna be solid too, no doubt.
Super hyped. My favorite part of RGT is finding old games being translated.

 

[TruxtonTatsujin TruxtonTatsujin]

Whatever, if you wanna give this guy blind praise don't let me stop you, but maybe you should at least check out some of his other work before you do so.

 

[UnDubious]

Y’all complaining about the translation like have you’ve actually played old PS1 games. Those were straight-up garbage translations half the time.
imo people are just freaking out because they saw “machine translation” and lost their minds.
I finished 1vierock’s partial english Panzer Bandit translation and it was totally fine.
This one’s gonna be solid too, no doubt.
Super hyped. My favorite part of RGT is finding old games being translated.

I've played dozens of PS1 RPGs and most of them read just fine. People jump on a few random lines and pretend the whole thing is garbage. With machine fan translations, you can't play 20 minutes without running across something crazy. Panzer Bandit is an action game where what little story exists doesn't matter. It's not comparable.

Post automatically merged: Oct 11, 2025

This ain't college and I'm not your professor, you can use google to find sources,

"I'm so used to being lazy that I'll make claims, post a random link that doesn't back it up, and then get angry when that's enough."
More importantly, there's been several quality PS1 fan translations already, and literally hundreds for various consoles, long before AI existed. Pretending it's impossible is absolute nonsense. Maybe those people on reddit think the translations are "better than fan translations" because they're talking about porn VNs, which probably don't carry the highest standards. Most completed fan translations for real games have been pretty good. I've been playing them forever.

Post automatically merged: Oct 11, 2025

Also, you haven't read the article about Crunchyroll you posted... it's literally about problems with bad machine-translated subs. It doesn't back up the idea that those are good at all.

 

[Remisse]

This ain't college and I'm not your professor, you can use google to find sources, but I'll do it this once for all the people who just talk/complain a lot and don't actually do the hard work like 1vierock is.
Anyone calling AI as slop makes them instantly irrelevant.

You make a claim, you provide sources to support that claim. It's always worked like that.

Almost all of the sources you provided are kinda irrelevant to what I've asked though, which is proving your claim that "most official translators" (your words, not mine) working on localization projects only require light editing. I honestly feel like you just wanted me to waste my time, but I'll entertain you nonetheless:

- https://www.reddit.com/r/visualnovels/comments/1itjthy/why_are_ml_translations_still_looked_down_upon_in/ (in 2025. . . on par with / sometimes surpass fan translations” when used with context; good for hobbyist perspective.)

Anecdotal ("LLMs good, trust me") and does not address your claim.

- https://www.reddit.com/r/visualnove...ude_3_sonnet_translation_sample_of_an_entire/ (note from reddit: I've tested quite a few LLMs over the last months and I am quite impressed at what Claude 3 Sonnet can do. It seems to get the context right 99% of the time, unlike ChatGPT4 which trips up quite often.)

Compares MTL with the Japanese sources without providing a baseline made by somebody who knows the language to prove how accurate those machine translations are. That means neither me nor you can verify how good the results are, and that's the entire point I've been trying to make. "It sounds good enough" is not a good benchmark and you know it. Doesn't address your claim either.

- https://www.reddit.com/r/JuJutsuKai...is_a_mistranslation_the_official_translation/ ( one user said a gpt 4 translation made more sense that the official)

Shows that the original translator was just terrible at their job. Nobody said humans are infallible. Doesn't address your claim.

- https://news.ycombinator.com/item?id=40219765 (I find ChatGPT (4) to have excellent Japanese skills already. It seems to produce more accurate and certainly more idiomatic translations than Google Translate does, and can even explain its own translations afterwards.)

More anecdotes, does not address your claim.

Now let's get to the only two decent sources you've provided:

• You conveniently posted the one article about the CrunchyRoll controversy that doesn't show more damning screenshots of the subs, in addition to it not even supporting your argument in the slightest. It seems the thing required more than just some light editing (but maybe your definition of "light" is vastly different from mine):
  • https://www.engadget.com/entertainm...ty-vendor-for-ai-subtitle-mess-145621606.html
  • https://www.theverge.com/ai-artific...bad-chatgpt-subtitles-on-its-new-anime-series
• The Japan Times article is exactly the kind of thing I asked for, but... that's all you've got? You said "most" translators were doing this, yet you only managed to find one source?

So, you only posted 1 valid source which ultimately is not even enough to validate your claim. Maybe try asking Grok for more relevant sources next time, and make sure you actually read through them before commenting again.

However, I think the real problem here is that there's a reading comprehension issue that's completely derailing the conversation. You desperately want to make it look like people here are against LLMs and machine translations in general, yet what we're saying is that vierock needs somebody knowledgeable to look at his work and make it actually decent. Like c'mon now, you can't possibly be looking at the screenshots he posted and tell me those sentences are well-written. Be for real for a minute.

Post automatically merged: Oct 11, 2025

Y’all complaining about the translation like have you’ve actually played old PS1 games. Those were straight-up garbage translations half the time.
imo people are just freaking out because they saw “machine translation” and lost their minds.
I finished 1vierock’s partial english Panzer Bandit translation and it was totally fine.
This one’s gonna be solid too, no doubt.
Super hyped. My favorite part of RGT is finding old games being translated.

I'm sorry but, are we currently living in the 90s? Shouldn't we strive to do a better job so we don't have to suffer through the same crap we got back then, now that we have better tools at our disposal? What kind of argument is this?

 

[Hikaru1057]

Guys, i think it's better to move the debate into another thread, i don't like it when it escalates into a flame war, make it civil though, i'm not used to toxicity and i hate it with a passion

 

[mighthyman]

Y’all complaining about the translation like have you’ve actually played old PS1 games. Those were straight-up garbage translations half the time.
imo people are just freaking out because they saw “machine translation” and lost their minds.
I finished 1vierock’s partial english Panzer Bandit translation and it was totally fine.
This one’s gonna be solid too, no doubt.
Super hyped. My favorite part of RGT is finding old games being translated.

100% they saw “machine translation” and lost their minds.

The screenshots of the game are 100% acceptable, if people want something better. Nothing is stopping them from doing it themselves.
Instead they harass people doing the hard work, and chase them out of the community like the guy who translated Shin Megami Tensei 1 for the psx.

If people the time to write essays, why don't they go learn mips reverse engineering and translated the oldpsx games. Because ITS WAY HARDER than the writing an essay or complaining. Duo lingo, lokalize, etc are all proving AI can do human level translation today in 2025. People will keep moving the goalposts on AI all day long.oh It's not human level, but human level translations suck too. Crunchyroll sucks blah blah blah, while the builder's keep on building.

Builders, you keep on building, We support. you Ignore the lazy loud complainers, who don't actually do any work.

Post automatically merged: Oct 12, 2025

You make a claim, you provide sources to support that claim. It's always worked like that.

Almost all of the sources you provided are kinda irrelevant to what I've asked though, which is proving your claim that "most official translators" (your words, not mine) working on localization projects only require light editing. I honestly feel like you just wanted me to waste my time, but I'll entertain you nonetheless:

Anecdotal ("LLMs good, trust me") and does not address your claim.

Compares MTL with the Japanese sources without providing a baseline made by somebody who knows the language to prove how accurate those machine translations are. That means neither me nor you can verify how good the results are, and that's the entire point I've been trying to make. "It sounds good enough" is not a good benchmark and you know it. Doesn't address your claim either.

Shows that the original translator was just terrible at their job. Nobody said humans are infallible. Doesn't address your claim.

More anecdotes, does not address your claim.

Now let's get to the only two decent sources you've provided:

• You conveniently posted the one article about the CrunchyRoll controversy that doesn't show more damning screenshots of the subs, in addition to it not even supporting your argument in the slightest. It seems the thing required more than just some light editing (but maybe your definition of "light" is vastly different from mine):
  • https://www.engadget.com/entertainm...ty-vendor-for-ai-subtitle-mess-145621606.html
  • https://www.theverge.com/ai-artific...bad-chatgpt-subtitles-on-its-new-anime-series
• The Japan Times article is exactly the kind of thing I asked for, but... that's all you've got? You said "most" translators were doing this, yet you only managed to find one source?

So, you only posted 1 valid source which ultimately is not even enough to validate your claim. Maybe try asking Grok for more relevant sources next time, and make sure you actually read through them before commenting again.

However, I think the real problem here is that there's a reading comprehension issue that's completely derailing the conversation. You desperately want to make it look like people here are against LLMs and machine translations in general, yet what we're saying is that vierock needs somebody knowledgeable to look at his work and make it actually decent. Like c'mon now, you can't possibly be looking at the screenshots he posted and tell me those sentences are well-written. Be for real for a minute.

Post automatically merged: Oct 11, 2025

I'm sorry but, are we currently living in the 90s? Shouldn't we strive to do a better job so we don't have to suffer through the same crap we got back then, now that we have better tools at our disposal? What kind of argument is this?

I'm not writing an academic thesis here. There is plenty of proof online.
I would absolutely say light editing indeed. There is a ton of evidence, that AI understands the context and nuance of japanese, and can even write context notes. But you discount reddit where people are actually doing experiments and showing their work. Your sources are engagdet and the verge? haven't you heard of the https://en.wikipedia.org/wiki/Gell-Mann_amnesia_effect .
People showing their experments and results on reddit are superior than your "sources".

Either way if you have a problem with this translation. No one is stopping you from making your own.

 

[Remisse]

100% they saw “machine translation” and lost their minds.

The screenshots of the game are 100% acceptable, if people want something better. Nothing is stopping them from doing it themselves.
Instead they harass people doing the hard work, and chase them out of the community like the guy who translated Shin Megami Tensei 1 for the psx.

If people the time to write essays, why don't they go learn mips reverse engineering and translated the oldpsx games. Because ITS WAY HARDER than the writing an essay or complaining. Duo lingo, lokalize, etc are all proving AI can do human level translation today in 2025. People will keep moving the goalposts on AI all day long.oh It's not human level, but human level translations suck too. Crunchyroll sucks blah blah blah, while the builder's keep on building.

I'm not writing an academic thesis here. There is plenty of proof online.
I would absolutely say light editing indeed. There is a ton of evidence, that AI understands the context and nuance of japanese, and can even write context notes.

You keep proving my point: you derail the discussion towards the quality of AI in general when nobody is even disputing the improvements of the most recent models, and towards irrelevant topics like telling people to learn hacking when 1) vierock's hacking is not even what we're criticizing, and 2) just as you like to criticize human-made translations 'cause they "suck too" without being a translator yourself, we're free to criticize AI translations however we want. And that's because you got nothing else to go on.
You dismiss human-made translations, yet you keep failing to provide any sorts of verifiable benchmarks on AI translations or anything at all that would substantiate your "light editing" claim.
Again, it's your job to prove your brittle arguments, certainly not mine lmao.

But you discount reddit where people are actually doing experiments and showing their work. Your sources are engagdet and the verge? haven't you heard of the https://en.wikipedia.org/wiki/Gell-Mann_amnesia_effect .
People showing their experments and results on reddit are superior than your "sources".

Either way if you have a problem with this translation. No one is stopping you from making your own.

I'm not sure you've noticed, but I've never even mentioned Reddit in the answer you're quoting. I've only attacked the contents of those posts by showing you that they're all irrelevant to the issue at hand, whether they come from Reddit or whatever other forum you're visiting.

And for the record, I'm not looking to drive anybody away from this hobby like you're claiming. vierock is doing well as a hacker, but to make these projects truly shine I feel he needs to form a team. I see this project has garnered quite a lot of interest from people in this forum, so I'm pretty sure he could find somebody if he posted a call for aid or something. He could still use AI translations as a starting point, and then leave it to a translator to fix all mistakes. It's not a hard concept to grasp, but to keep telling him that this is great and that there's nothing he should change about his process means that perhaps you don't want him to improve. But I do, and it really saddens me to see all this just because you're not able to see the limits of current-day AI.

 

[TruxtonTatsujin TruxtonTatsujin]

You keep proving my point: you derail the discussion towards the quality of AI in general when nobody is even disputing the improvements of the most recent models, and towards irrelevant topics like telling people to learn hacking when 1) vierock's hacking is not even what we're criticizing, and 2) just as you like to criticize human-made translations 'cause they "suck too" without being a translator yourself, we're free to criticize AI translations however we want. And that's because you got nothing else to go on.
You dismiss human-made translations, yet you keep failing to provide any sorts of verifiable benchmarks on AI translations or anything at all that would substantiate your "light editing" claim.
Again, it's your job to prove your brittle arguments, certainly not mine lmao.

I'm not sure you've noticed, but I've never even mentioned Reddit in the answer you're quoting. I've only attacked the contents of those posts by showing you that they're all irrelevant to the issue at hand, whether they come from Reddit or whatever other forum you're visiting.

And for the record, I'm not looking to drive anybody away from this hobby like you're claiming. vierock is doing well as a hacker, but to make these projects truly shine I feel he needs to form a team. I see this project has garnered quite a lot of interest from people in this forum, so I'm pretty sure he could find somebody if he posted a call for aid or something. He could still use AI translations as a starting point, and then leave it to a translator to fix all mistakes. It's not a hard concept to grasp, but to keep telling him that this is great and that there's nothing he should change about his process means that perhaps you don't want him to improve. But I do, and it really saddens me to see all this just because you're not able to see the limits of current-day AI.

You can't reason with AI tech bros, just let it be. They're not worth your time.

 

[Remisse]

You can't reason with AI tech bros, just let it be. They're not worth your time.

He may be hopeless, but the goal of my comments is to inform other people about the limits of LLMs, hoping they can see that, despite AI being potentially great when used properly, you still need knowledgeable humans to be able to verify the validity of their output. Or else you end up with rather low-quality projects like this one, which is super unfortunate considering it could become so much more with actual effort.

There was no need for him to resort to underhand tactics like claiming I or other people want vierock to quit. That's absolutely false, but I guess that's what he does when he realizes he's lost an argument.

 

[LordRay589]

Hi, just to say that I've taken over,that the creator gave up in 2010.

View attachment 112973View attachment 112977View attachment 112975View attachment 112976​

I root for this game to be translated and hope to see their badass skills, bro.