translators(mki-多语言翻译)

Ever wanted to run your English prompt through Japanese or Chinese because the model you're using visibly prefers its native language - or translate a model card's recommended tags into something you can actually read? That's this node. translators(mki-多语言翻译) wraps the Python translators library, which drives dozens of free online translation services - Google, Bing, DeepL, Baidu, Alibaba and more - with zero API keys and zero accounts. The name is a small lie in the good way: it doesn't call one API, it keeps an array of them to fall back on.

Why you'd reach for it. Prompt translation is a genuinely common workflow, not a gimmick. The KB's prompt-engineering essay notes the community increasingly routes natural-language intents through translation on the fly; this is that idea done cheaply, locally, and without an LLM API bill. You wire the output STRING straight into a CLIP Text Encode node and get a prompt your model actually groks. It's equally handy for translating negative prompts or dataset captions before batch runs.

How it works. The node sets the translators_default_region environment variable to EN, then calls the library's translate_text() on a background thread with a 10-second timeout. With from_language set to auto the service detects the source language itself. Each of the 39 services speaks a slightly different dialect of language codes, so the pack ships a mapping that normalizes codes per translator - which is why the dropdowns show "english(en)" style entries instead of raw codes. The if_use_preacceleration toggle runs the library's preaccelerate_and_speedtest() once per node instance: it pre-fetches the cookies and tokens a service wants, making later calls faster at the cost of a slow first run.

The inputs that matter:

• query_text - the text to translate (multiline).
• translator - one of 39 services; the default is alibaba, and if a service starts rate-limiting you, switch to Google or Bing and it usually just works.
• from_language (auto + 112) and to_language (112, default English).
• if_use_preacceleration - worth flipping on if you're doing many calls in a row.

The output is a single STRING with the translated text.

Gotchas, and they're real ones. This is a network node; the README's entire requirement is that your connection works. If a service hangs, the 10-second timeout kicks in and you get your original text back unchanged - the node logs a warning to the console, so check there before assuming translation ran. Free services rate-limit aggressively and some block datacenter IPs, so "it stopped translating" almost always means "switch translator," not "restart ComfyUI." If a service errors with Unsupported or same, the node falls back to returning your input text - usually a sign the language pair isn't supported by that particular service.

Install. The pack's requirements.txt pulls in translators>=6.0.0, langdetect, and sentencepiece:

cd ComfyUI/custom_nodes
git clone https://github.com/MakkiShizu/ComfyUI-MakkiTools
cd ComfyUI-MakkiTools
..\..\..\python_embeded\python.exe -m pip install -r requirements.txt   # Windows portable
pip install -r requirements.txt                                          # venv/conda

Or search "ComfyUI-MakkiTools" in ComfyUI Manager, then restart ComfyUI so the new packages are importable.