Hey all,

We've just released a new online manga translation tool, which is available here.

For every page uploaded, we use machine learning to detect the text and provide the translator with inline suggestions to dramatically speed up the translation. Once translation is complete, you can click “export” to perform the actual typesetting, and receive a finalised manga page in English ready to be shared with the community. We believe this can save a lot of time and effort, resulting in more manga for all of us 😀

Who would like to try it out? we would be very happy to hear your feedback.

That's pretty awesome and helpful! Thanks for the translation tool.
Do you have github page for this app?
For suggestion that maybe you can add "Suggest better translation" to improve the translation, then maybe alternative sentence translation like what google translate do.

Thanks 🙂

No GitHub unfortunately, the tool uses heavily our company's internal services so we can't make it open source. However, it is and will be open for the scanlation community for free, forever.

You can actually edit the translation right now, it is a textbox which you can freely edit (also resize, add new textbooks and delete existing ones) before clicking on "export". Our algorithms will learn & improve from the corrected translation so they can become better with time.

What other features would be helpful for you?

There's an option to log in with email but where's the button to actually sign up first so that you can log in? 😕

Sorry about that. It’s in the works, should be up by end of the week (along with ePub import and export support).

In the mean time, if you want you can send me a private message with your email and we’ll create an account for you and send the credentials over.

Alternatively, I can update you when the new release is up, whatever works best for you.

Just tried it out with a few raw pages of Poco's Udon World and it works wonderfully! I mean, this manga is more to slangs and dialects so seeing that it was able to translate quite alright is really great. (I couldn't provide photos bcs I don't know how;; 😉

Machine translations aside, I personally think it's actually a great aid for learning Japanese, haha. Also, I like the function where we can export the translations to a spreadsheet file. It's easy to navigate and use, too. I haven't found anything to suggest yet but overall I'm liking it so far.

Thank you so much for this and good work! I really like it. I'm sure it'll be useful to be used for scanlations group as well. 😀 😀 😀

Wow, I'm excited a translating tool is out. Please make sure to get registration open because I'm going to definitely try the site out 🙂

Currently getting my life together~ (unsuccessfully)

that is fantastic!

I'd like a feature to manually edit the bubble mask so you're not limited to just a rectangle underneath the text box.
Or perhaps an option to disable the bubble mask entirely, so a user could manually do it in Photoshop later (possibly with some kind of action).

I guess there's also currently no way to save to PSD, so that would be helpful.

Also, As much as this is a translation tool, it also seems like it'd be pretty nice to skip the machine translation and just use it for speedier typesetting, so I would like there to be some way to import a script alongside a set of pages and have it automatically place the lines wherever it finds OCR text.

Edit: Another feature request would be some kind of selection tool that you could use to manually select any bubbles or text that it doesn't find automatically, and it could re-do OCR on those specific areas.

Visit my scanning blog: Jammin' Scans

::End of Transmission::

I am extremely impressed with the accuracy at finding Japanese text and placing text boxes on top of it. However, I am less impressed with the machine translation part. It doesn't speed anything up, unfortunately; it is actually more a distraction than anything else.

Now, I don't know if I'm doing something wrong, but when I try to enter my own text in the bubbles, it is entered backwards (right to left, bottom to top)? Is that a bug or a feature?

But overall, with some incremental improvements, this could be revolutionary in scanlation, especially if you can combine it with an online reader where translators can just come in, add the text, and it shows up immediately. I.e. scanlation as we know it would be dead.

Active translations list
Completed translations list
Dropped translations

So, the idea that translator can " can combine it with an online reader where translators can just come in, add the text, and it shows up immediately" is actually the backbone of raffmanga before things got a little to out of hand. So this has been done. The problem is the raw source isn't always in the best quality....

But the translation tool is a good idea.

Thank you so much 🙂 registration with email & password coming very soon! In the mean time, you can register with Facebook or Google account if that helps.

Quote from vigorousjammer

I'd like a feature to manually edit the bubble mask so you're not limited to just a rectangle underneath the text box.
Or perhaps an option to disable the bubble mask entirely, so a user could manually do it in Photoshop later (possibly with some kind of action).

You can actually resize, move, delete and add bubbles! to resize just drag from the bottom of the bubble, to move it around click and hold its borders. We will also have multiple shapes of textbooks in future so it'll be easier to fit.

Quote from vigorousjammer

I guess there's also currently no way to save to PSD, so that would be helpful.

PSD support is in the works, however this format is a little more complicated than ePub so it will take us a little time to get there. It would actually really help us if you can how would you like such a PSD to be structured? do you want the translated text in asseverate layer? maybe have each bubble in its own layer?

Quote from vigorousjammer

Also, As much as this is a translation tool, it also seems like it'd be pretty nice to skip the machine translation and just use it for speedier typesetting, so I would like there to be some way to import a script alongside a set of pages and have it automatically place the lines wherever it finds OCR text.

Didn't think of that, very cool feature. How would you like to import the script? excel file or cvv works for you?

Quote from vigorousjammer

Edit: Another feature request would be some kind of selection tool that you could use to manually select any bubbles or text that it doesn't find automatically, and it could re-do OCR on those specific areas.

Also an awesome feature. We will add it hopefully in next week's release, will keep you posted. I think we can have two tools, one that lets you mark an area to trigger OCR and translation, and another that lets you tap somewhere in the bubble and detect its boundaries automatically. On it!

Quote from cmertb

I am extremely impressed with the accuracy at finding Japanese text and placing text boxes on top of it. However, I am less impressed with the machine translation part. It doesn't speed anything up, unfortunately; it is actually more a distraction than anything else.

Now, I don't know if I'm doing something wrong, but when I try to enter my own text in the bubbles, it is entered backwards (right to left, bottom to top)? Is that a bug or a feature?

Good question, sounds like a bug. I'll PM you to get more info. Translation will take time to improve, but each time you use it and correct the original suggestion our optimisation layer will become better! Also, would you like to have the ability to disable the translation suggestions? Would you want us to leave empty bubbles as a placeholder?

Quote from cmertb

But overall, with some incremental improvements, this could be revolutionary in scanlation, especially if you can combine it with an online reader where translators can just come in, add the text, and it shows up immediately. I.e. scanlation as we know it would be dead.

Thank you so much, our team's Slack is filled with many happy emojis right now 🙂 Export to a reader is in final stages of development, it will hopefully be available in this week's version - if not, it'll be in the one after that tops.

Quote from F_J

So, the idea that translator can " can combine it with an online reader where translators can just come in, add the text, and it shows up immediately" is actually the backbone of raffmanga before things got a little to out of hand. So this has been done. The problem is the raw source isn't always in the best quality....

But the translation tool is a good idea.

Thanks F_J! I didn't know about raffmanga

Quote from F_J

So, the idea that translator can " can combine it with an online reader where translators can just come in, add the text, and it shows up immediately" is actually the backbone of raffmanga before things got a little to out of hand. So this has been done. The problem is the raw source isn't always in the best quality....

I remember seeing something like that. I think I posted some feedback that never got addressed, and then I never heard about that project again.

Quote from Nadavuu

Translation will take time to improve, but each time you use it and correct the original suggestion our optimisation layer will become better!

This sounds good in theory, but I'm extremely skeptical that this feature will ever work for Japanese dialogue (which is what manga is). I have tried using a translation memory database in fansubs (gtk provided by Google), and the results were more frustrating than anything. The problem with Japanese, which sets it apart from most IE languages, is that it's so highly context dependent that exactly the same line might be translated very differently depending on where it appears. Thus, even when you get a perfect match, it's still not helpful most of the time. Well, we'll see -- there might have been a lot of progress over the years.

I'll also post some feature suggestions that would be indispensable to scanlators a bit later.

Active translations list
Completed translations list
Dropped translations

Quote from cmertb

This sounds good in theory, but I'm extremely skeptical that this feature will ever work for Japanese dialogue (which is what manga is). I have tried using a translation memory database in fansubs (gtk provided by Google), and the results were more frustrating than anything. The problem with Japanese, which sets it apart from most IE languages, is that it's so highly context dependent that exactly the same line might be translated very differently depending on where it appears. Thus, even when you get a perfect match, it's still not helpful most of the time. Well, we'll see -- there might have been a lot of progress over the years.

Totally agree - manga is probably one of the hardest machine translation problems out there due to all of the above, and also because it is a format that encourages brevity and uses a combination of text and imaging to advance the narrative. But we like hard problems 🙂

Essentially, our approach is firstly to "learn" how to optimise the translation per specific manga series and retrain our algo when we make mistakes. Also, we use computer vision to provide context for the machine translation algorithms, so for example translation may change for male or female, or if the scene depicted is a battle or a love story. We try utilise all the hints we can find to firstly create the context which will in turn improve the translation, and then do another pass to optimise and adjust for the specific manga, with all it's potential years of backstory, inside jokes and specific usage of language. It'll be very hard to reach a perfect translation, but this approach shows major improvements so far, and we hope that as the suggestions get better they become more useful for users.