Text Translation V2 (Full)

Text Translation V2 (Full) is the kitchen-sink version of the pack's translator. Where plain V2 gives you three engines and eleven languages, Full hands you 37 translators - Alibaba, Apertium, Argos, Baidu, Bing, Caiyun, DeepL, Google, Yandex, Papago, Youdao, and dozens more - plus around 48 source languages and 47 target languages, including niche ones like 文言文 (Classical Chinese) and 壮语 (Zhuang). Same translators library underneath, same no-key, no-account, no-model setup. If you need a specific engine or a specific language pair, this is the only node in the pack that can do it.

The one feature V2 doesn't have

On top of everything V2 offers, Full adds a recognition_switch, on by default. Before translating, it runs your text past a regex that checks for non-English characters. If your text is already all-English - letters, digits, common punctuation, emoji - it skips translation entirely and prints a little note to the console ("No non-English characters detected, skipping translation."). That's a genuinely thoughtful touch for a prompt workflow: run your whole prompt through it and English fragments don't waste an API round-trip.

The catch is the wording. It's "has non-English characters," not "is Chinese." That's almost always what you want, but it means a prompt with a stray accented character still gets sent off for translation. Don't treat the recognition switch as language detection - treat it as a "skip if it's English enough" filter.

Inputs and outputs

• recognition_switch - skip translation when input is already English. On by default.
• trans_switch - the master on/off. On translates, off passes text through.
• trans_text - the multiline text to translate.
• translator - 37 engines. Note the default is the first item in the list, which is Alibaba, not Bing.
• source_language / target_language - the big dropdowns. Languages display as 中文(zh) and the node parses out the code.

Output is a single string, ready for a CLIP Text Encode or any downstream text node. Like V2, extra options hide behind a front-end "Show / Hide button" so the node doesn't open as a wall of dropdowns.

The honest caveat

Not every engine supports every language pair - some of those 37 are region-locked, some barely maintain their language lists, a few are effectively dead. The README says it straight: test before you trust a given engine + language combination, and cross-check against the translators library's documentation for what each backend actually supports. For most prompt-to-encoder work you'll never need this node; plain V2's Bing/Google/Alibaba covers the realistic cases. Reach for Full when you specifically want DeepL quality, a language V2 doesn't list, or you're translating something other than a diffusion prompt.

Installing

It lives in the pictorialink/ComfyUI-Text-Translation pack with the rest. ComfyUI Manager: search ComfyUI-Text-Translation. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/pictorialink/ComfyUI-Text-Translation
cd ComfyUI-Text-Translation
pip install -r requirements.txt

Restart ComfyUI. The only dependency is translators; there are no model downloads and nothing to configure. Old pre-May-2024 installs of this pack are incompatible with the current one - delete the old folder first. And if you ever see the Value not in list validation error, that's stale widget state after an update: delete the node and re-add it.