Mondblut
-
Posts
14 -
Joined
-
Last visited
Content Type
Profiles
Forums
Blogs
Events
Posts posted by Mondblut
-
-
-
On 11/11/2023 at 11:10 AM, shiny5478 said:
The problem is after telling people it's an MTL people will not give you tools. It's that sort of community, unfortunately. Super sekrit club
I learned my lesson. Being honest and open is never getting you what you need and lying is always more effective to achieve one's goals. Sadly I'm an honest person.
Either way: I've released it a while ago as a spreadsheet.
I was sadly unable to decrypt the files and making a patch. I'm currently working on a Java based VN overlay subtitle app which works like Sugoi Japanese Translator as an overlay textbox loading the text directly from the spreadsheet. It should bypass the issue that a patch is not possible. -
-
1 hour ago, Crsky said:
GARbro can't resolve filenames from a XP3 file. Because what is stored in the XP3 file is a hash value of the original filename, not the filename itself. GARbro cannot reverse the hash.
To get the file names, you need to use KrkrDump, a tool that extracts files read by the game engine in real time. However, please note that it cannot extract all files at once.
Seems to work. So I basically need to play the game and it will dump the script files eventually? But how do I put them back into the game after editing the script?
Moreover: is there a way to run the game via locale emulator while using the dumper? -
So I managed to extract the xp3 files, and they seem to be decrypted now. But the files are all without any extension.
In the Garbro viewer I see Scenario_01.txt and so on but how am I to extract these txt files? Where are they even? They aren't in the list on the left.
-
1 hour ago, Crsky said:
Thanks a lot. I will give it a try.
-
I've tried Translator++ so far, and it extracts the files, but they have no file extension and it can't seemingly detect them. Probably they are encrypted? I tried KrKrExtract, but it doesn't work on 64 bit systems. I also tried Garbro, however I have no idea if it decrypts them correctly or not. And how do I open them and work on the script?
-
Are there any extraction and repacking tools for Mashiro Iro Symphony or any other early Palette games? I only found this:
https://web.archive.org/web/20190309075007/http://www.geocities.jp/hoku_hoshi/souko/souko078.html
But apparently it's just for extraction. -
10 hours ago, Clephas said:
Essentially, a decent translation of a scene requires the translator to comprehend the entirety of the context of the scene up until the line in question, which is where MTL seems to fail the most drastically.
Which is exactly why ChatGPT is such an evolutionary step. It analyzes the entire block you submit before it attempts to translate. The more text you provide, the more context it gets. Essentially: if the processing power increases and entire VN scripts can be fed into such an AI, the result would be not worse than how a human native speaker being fluent in both languages would translate. Sadly right now ChatGPT has a limit of how much it lets you process, so you offer it only limited context, but it's context no less.
As an experiment I let ChatGPT edit a block of 50 lines of a bad MTL which had the characters not use correct personal pronouns and names instead of "my," "I" or "your." I ordered it to infer the pronouns correctly from the context of the text. And It surprisingly did so flawlessly. Even if it had no speakers listed (so basically without Person A: "...." Person B: "...."). If the speakers are listed next to the lines it would do so even more reliably. Personally I think this technology is fascinating from a translation context and we might see it develop even further in the years to come.
-
I need to extract the character names and the dialogue separately. So in a second run I extract JUST the character names. However it omits repeated character names. If character A says something and he says again something, right after, his character name doesn't show up as a new extracted line in the extractor window.
It looks like this:
It's not an extension causing it and I didn't set "filter repetition" in the settings. Could it be that Textractor by nature only adds a new extracted line to the log if it's different from the last line? I didn't notice that behaviour with actual dialogue but then again dialogue that repeats itself exactly is extremely rare to begin with, unlike names, like in this case here. It could of course also be the character names not refreshing as a new string when the character says something again right after. I don't know. Any idea? Any idea how to fix that? -
2 hours ago, HataVNI said:
It's not a violation of the rules, but I'm not going to promote that project or share it on Twitter or cover it on the blog. Otherwise we would need to cover all the SVN (Sugoi Visual Novels) projects as well and there are countless of those whose quality is unverified. Also it's simply against our translator/localizer belief (some appropriate localization is ok, memelation is not.)
And I would never ask of this or any site to promote this MTL I'm working on. It's just something I'm working on for myself to be able play something that will probably never see an actual translation. And since I don't want to keep this just to myself I decided to share this with others. Also the thread was not meant as a project announcement or anything. I just wanted to know if someone knows how to work with Purple Software files. Probably I shouldn't have even mentioned that I'm working on an MTL of the eroge.
Also: the Sugoi VN MTLs read like garbage. The only one I tried is Maitetsu Last Run and quite honestly it's better to just use Sugoi Japanese Translator with DeepL and even that is unreadable.
2 hours ago, HataVNI said:MTL-tech isn't there yet, so it shouldn't be encouraged to use. It also puts legit people (not just the memelators you so hate) out of jobs, because employers like to cut as many corners as they can to keep costs low - which is also not great.
It's very close though. I think in 2 or 3 years we will have Chat AIs that can read and process an entire VN/eroge at once, therefore having the context of each line as a human translator would. Once this point is reached localizers will be obsolete. Given what kind of ilk works in that profession nowadays, I can't wait for that day to come soon enough.
The only ones I feel sorry for are passionate fan translators who will be discouraged by this technological evolution. It's actually the reason why I'm considering to just release this "project" as a spreadsheet and nothing else, so that no one will be discouraged to do an actual TL of that or any other eroge in the future. It's actually a moral dilemma I've been pondering recently. -
1 hour ago, HataVNI said:
ChatGPT doesn't do ero, so as soon this happens, it will fail. There's possibility to jailbreak that stuff, but this only works temporarily. Learning Japanese is definitely important to translate, so we can't really condone this. We can't stop you either though, so we can just appeal that you at least find a TLC to look over this before you prepare a release.
Hello Mondblut, welcome to Fuwa.ChatGPT 3 which I use for my case does translate h-scenes, from 3.5 onward it doesn't. There's an example screenshot of a properly ChatGPT translated h-scene in the Reddit thread.
Either way, is it a violation of this forum's rules? If so I apologize and don't mind if you delete my thread. I just wanted to do this for myself and eventually decided to share it with others once finished. It's not some official project or anything.
Either way, it seems Purple Software's .cpz files seem to have some encryption. Looks like it will stay a spreadsheet.
-
I started to do a ChatGPT based machine translation of Mirai Nostalgia. Due to the lack of any tools to open and re-pack .cpz files I've worked with textractor and a simple spreadsheet for now. However the more time I've put into this project, even slightly started editing the text, the more I realized that it would be wise to rather do an actual patch than just a spreadsheet kind of translation walkthrough.
Does anyone have any knowledge about Purple Software games? Is there any tool that lets you extract and re-pack .cpz files?
The progress so far btw.
Common route: 100%Iori route: 100%
Uta route: 80% (probably)
Some proof of how good (or bad) the TL is for anyone interested:
https://www.reddit.com/r/visualnovels/comments/1775wlw/mirai_nostalgia_unedited_chatgpt_translation/
Edit: the tool (https://github.com/xmoezzz/CMVS-Engine) linked here in the forum does nothing when dropping the .cpz files into the exe and executing the exe via cmd doesn't do anything either. It's a bit frustrating that there's no command documentation on how to use the tool, so I'm not sure if I'm doing it right or not.
AstralAir no Shiroki Towa Partial Patch Released
in Visual Novel Talk
Posted · Edited by Mondblut
Watched a youtube video let's play of the prologue and it seems rather decent. It seems it has been heavily edited since the initial release, but is the entire game like this or are the routes still a train wreck? Has anyone read the entirety of it?
https://youtube.com/watch?v=gSQN90f4JTQ
Edit: does anyone still have the patch saved on a hard drive and can re-upload it? Considering that it's even worth the effort.