Japanese Romaji Converter (Advanced)

JapaneseRomajiConverterAdvanced is the same idea as the standard Japanese Romaji Converter from the same pack - turn Japanese text into Hepburn romaji so English-based video models like LTX-2.3 can actually pronounce it - with two extras bolted on: a mode that converts all Japanese instead of just quoted dialogue, and a passthrough of your original prompt so you can A/B what went in against what came out.

The realistic split between the two nodes: use the standard one for the common case (dialogue inside quotes, which is what 95% of prompts need), and reach for the Advanced when your workflow drops Japanese outside of quotes - scene notes, text overlays, a title card says: ... line - or when you're iterating and want to eyeball what the converter actually changed.

What's actually different

Same two engines as the standard node - fugashi (default, recommended) with its UniDic pronunciation-based conversion and smart part-of-speech spacing, or pykakasi as the lightweight fallback. Same prompt input. The new knobs:

• conversion_mode (COMBO, default quotes_only) - quotes_only behaves identically to the standard node: Japanese inside "..." gets romaji-ized, first letter capitalized, English everywhere else untouched. all_japanese instead converts every Japanese run in the whole prompt, quoted or not. Handy if your Japanese isn't always in quotes - the tradeoff is that it'll happily romanize Japanese scene directions too, which can read oddly but does feed the model pronounceable text.
• remove_english_translations (BOOLEAN, default true) - same blunt (parenthetical) stripper as the standard node. Be aware it removes any parenthesized text containing a Latin letter, not just English translations, so (4K) or (Tokyo) will vanish too.

Where the standard node returns one string, this one returns two: converted_prompt (the romaji version) and original_prompt (the untouched input). The second output is pure debug convenience - drop it into a Show Text node and you can diff before/after while you dial in your pipeline. Once it's working, you can just ignore that output; it's a free wire.

Installing it

No special steps - it ships in the same pack. ComfyUI Manager: search "ComfyUI Japanese Romaji Converter" and install. Manual:

cd ComfyUI/custom_nodes/
git clone https://github.com/hirokazu/ComfyUI-JapaneseRomaji
pip install fugashi unidic-lite

Then restart ComfyUI. The Advanced node appears alongside the standard one under LTX-Video/Prompt. No models to download, no GPU cost - it's a text transform. Like the standard node, fugashi is the installed default and pykakasi needs a separate pip install pykakasi if you want that engine as a fallback.

The honest take

This node exists for people who've outgrown the standard one, and if you're a beginner you can skip it entirely - the base node covers the main use case. Where it earns its keep: you have a prompt-generation pipeline (the README's LTX-2.3 flow runs scenario → romaji → LTX formatting) that emits Japanese outside quotes, or you want the before/after comparison while you figure out how aggressive to be with translation removal. One quirk to remember: the first-letter capitalization on quoted text is hardcoded in both nodes, so if you don't want "Koohii..." starting with a capital, there's no toggle for it - edit after the fact. Small price for a node that finally makes your Japanese dialogue come out of LTX sounding like Japanese.