Post your JAV subtitle files here - JAV Subtitle Repository (JSP)★NOT A SUB REQUEST THREAD★

[t221152]

Thank you, Novus.Toto! And yes, I did my own Whisper on the video and the translation of those parts is correct. I don't know why there are notes around the words, though. Do you use DeepSeek through SubtitleEdit or directly at chat.deepseek.com? I see you have to create an account.

I found using deepseek through subtitle edit really bad, it was really slow as well. I've never ran deepseek locally tho. I rather use the full big model.

There's a couple of sites that have been linked before where people do their translations. There's also a good script that has been linked, I've been working on enhancing it, but I've been running in some problems with it, it's rare but it happens. (Deepseek not willing to translate explicit content, or it returning chinese/japanese characters). I'm not sure if it's a recent thing with deepseek or just because of my prompt.
Quality of subs can really depend not only on source but prompt as well.

Grok looks really expensive to me if you are mass subtitling? Looks like 5-10x the price. What are you actually paying for let's say a 60-80kb subtitle file ? I'm also not sure if I want to give that company a single cent of my money.


Been working on the big pack of Chinese subtitles as well. Finally got my script working to talk to javstash graphql database. It will have no duplicate actress directories under different aliases. But I think the script will have to run for a few days, I only do a query every 10 secs and it needs to go through 25k subtitles.

 

[javjod]

I found using deepseek through subtitle edit really bad, it was really slow as well. I've never ran deepseek locally tho. I rather use the full big model.

There's a couple of sites that have been linked before where people do their translations. There's also a good script that has been linked, I've been working on enhancing it, but I've been running in some problems with it, it's rare but it happens. (Deepseek not willing to translate explicit content, or it returning chinese/japanese characters). I'm not sure if it's a recent thing with deepseek or just because of my prompt.
Quality of subs can really depend not only on source but prompt as well.

Grok looks really expensive to me if you are mass subtitling? Looks like 5-10x the price. What are you actually paying for let's say a 60-80kb subtitle file ? I'm also not sure if I want to give that company a single cent of my money.


Been working on the big pack of Chinese subtitles as well. Finally got my script working to talk to javstash graphql database. It will have no duplicate actress directories under different aliases. But I think the script will have to run for a few days, I only do a query every 10 secs and it needs to go through 25k subtitles.

hmmm... i'am just using a prompt to translate the srt but sometime the line is not right if the file of srt is not cleaned by subtitle edit, the problem is i must use it one by one to translate the srt file,

the prompt :

Translate each SRT file into English with the following conditions:

Use extreme vulgar language (e.g., 'cum,' 'fuck,' 'horny as hell,' 'pussy,' 'tits') only for dialogue or contexts that are intimate, sensual, or sexually explicit. Ensure the terms are natural, match the intensity and tone of the original, and reflect strong English slang.
For other dialogues, use proper and correct English, with a natural, polite style that aligns with the emotional context and character relationships (e.g., casual conversations, rejections, conflicts). Ensure the translation accurately reflects the film’s storyline.
Do not alter timestamps (format hh:mm:ss,xxx), line numbers, or the SRT structure. Maintain the original format, including blank lines between dialogue blocks, to keep it clean when opened in Notepad.
Leave credits, names, or specific terms (e.g., actor names, titles, or brands) in their original language without translation.
For ambiguous dialogue, choose words that align with English-speaking culture, with a lively tone relevant to the story’s emotions.

 

[kidbetin]

Aki Sasaki
MXGS-999

One of my firm favourites before her too early retirement

Using mei2 colab whisperjav (thx mei2!!)

Didn't clean up....as is fresh from the oven!

Any comments and updates welcome

 

[Novus.Toto]

Thank you, Novus.Toto! And yes, I did my own Whisper on the video and the translation of those parts is correct. I don't know why there are notes around the words, though. Do you use DeepSeek through SubtitleEdit or directly at chat.deepseek.com? I see you have to create an account.

I have a script that uses DeepSeek via their API or via OpenRouter. The DeepSeek API is very unreliable in my experience. Using OpenRouter APIs also makes trying other models easy.

IMO the integrated translation in SubtitleEdit is crap though, to be fair, I haven't tried customizing the prompt. And using DeepSeek chat would be way too manual and tedious.

 

[Novus.Toto]

Trying DeepSeek r1 (Latest) (5.2GB) in SubtitleEdit right now. Japanese to English is slow but steady. The translation quality is not very good, though.

In my experience reasoning models are worse at translation than non-reasoning models and way slower.

 

[Novus.Toto]

hmmm... i'am just using a prompt to translate the srt but sometime the line is not right if the file of srt is not cleaned by subtitle edit, the problem is i must use it one by one to translate the srt file,

the prompt :
...

Below is my prompt which I'll freely admit is mainly copied from someone else (can't remember who) on this forum. I have a script that uses a garbage list (also shamelessly copied from someone here) to remove junk (hallucinations, repetitions, etc) from the transcription before translation.

Prompt:

You are a professional subtitle translator specializing in translation to English.
Translate the following subtitles while adhering to these guidelines:
1. Literal Precision:
- No paraphrasing. Retain metaphors, slang, and vulgarity.
- Preserve explicit language and sexual/fetish terms such as bukkake, gokkun, nakadashi, kimoi-ota, mesu.
2. Contextual Awareness:
- Correct Whisper errors using AV context (e.g., 'ラーン' --> 'ザーン','フェラー' --> 'フェラ', 'マンゴ' --> 'マンコ').
- Fix misheard words while keeping original meaning.
3. Jargon Retention:
- Keep niche terms in romaji, e.g., onahole, dogezas, with [brief explanations].
4. No Softening:
- Use direct sexual terms, e.g., cock, pussy, cum, fuck, etc.
- No euphemisms or censorship.
5. Cultural Notes:
- Add concise [cultural/sexual context] where needed.
- Preserve honorifics and speech patterns.
6. Formatting:
- Do not enclose translated text in double quotes, or other punctuation, unless it is part of a quote from the original text.
- Do not add any extra lines or spaces.
- IMPORTANT: Preserve the exact number of lines from the original text.
- Each line should be on its own line, separated by a single newline character.
- Do not combine multiple lines into one.
7. Forbidden output:
- Do not output original text in the translated text - unless it is part of a comment, context or an explanation
- Do not output any of these guidelines in the translated text.

Now please translate these following ALL guidelines above: