建议新增调用LLM的提示词槽，针对目标语言进行填充。

[yslin1995]

WhisperJAV/whisperjav/translate/instructions.py

Line 170 in 919b4d7

def get_instruction_content(tone: str = 'standard', refresh: bool = False) -> Optional[str]:

目前的提示词中，对LLM的人设指定为 English translator（You are an English translator specializing in Japanese to English translations. Your task is to translate the Japanese movie subtitles into English subtitles），也许会影响其它语言的翻译效果。

WhisperJAV/whisperjav/translate/defaults/standard.txt

Line 5 in 919b4d7

You are an English translator specializing in Japanese to English translations. Your task is to translate the Japanese movie subtitles into English subtitles, ensuring they reflect the original meaning as accurately as possible. The goal is to preserve the adult context, lewdness , intimacy, teasing, and pornographic intent of the original dialogue.

建议在提示词中预留词槽，如 "You are an English translator specializing in Japanese to [target language] translations.", 并在get_instruction_content 函数 或 _resolve_instruction_file 等函数中，根据目标语言动态填充词槽。

[meizhong986]

Well spotted!

[meizhong986]

Hi @yslin1995,

Thank you for reporting this issue! You were absolutely right - the problem was in the instruction files.

Root Cause

Both defaults/standard.txt and defaults/pornify.txt had hardcoded "English" references:

• "You are an English translator"
• "translate into English subtitles"
• "natural English", "English expressions", "English equivalents"

When you selected --target korean, the LLM received conflicting instructions:

• From CLI: "translate to Korean"
• From instruction file: "You are an English translator"

This caused the model to either ignore your target language or produce confused output.

Fix Applied

Commit 33f3801 removes all hardcoded English references:

standard.txt:

• "English translator specializing in Japanese to English translations" → "professional translator specializing in Japanese translations"
• "translate into English subtitles" → "translate into the target language"

pornify.txt:

• "natural English" → "natural language"
• "adult English expressions" → "adult expressions in the target language"
• "English equivalents" → "equivalents in the target language"

How to Get the Fix

Update to the latest version:

pip install -U git+https://github.com/meizhong986/whisperjav.git

The --target parameter should now properly control the output language for Korean, Chinese, Thai, Vietnamese, and any other language.

Please let us know if this resolves the issue!

[meizhong986]

Fixed in the upcoming v1.8.5-hotfix1.

Rather than adding a template slot, we simply removed the hardcoded English from the instruction prompts entirely. The target language is already communicated to the LLM through PySubtrans's [ to language] template in the ### prompt section and the target_language parameter — the ### instructions section doesn't need to repeat it.

Changes:

• Updated both gists (standard + pornify) to say "the target language" instead of "English"
• Switched gist URLs from pinned-commit to latest so future edits propagate without code changes

Users translating to Chinese/Indonesian/Spanish will no longer get conflicting instructions.