Argos Translate Text Node

Every other translator in this pack - Google, Deep Translator, ChatGLM - needs a live network call to somebody else's server. ArgosTranslateTextNode is the odd one out: it's built on Argos Translate, an open-source neural machine translation library that runs entirely offline once the language models it needs are downloaded. No API key, no rate limits, no dependency on whether Google's translate endpoint happens to be having a bad day.

That trade-off cuts both ways, and it's worth understanding before you reach for this one specifically.

Why you'd reach for it

If you're running ComfyUI on a machine with no internet access (an air-gapped box, a locked-down network, a server where outbound calls are blocked), Argos Translate is genuinely your only option among the translator nodes in this pack - googletrans and deep-translator both need to phone out, and the ChatGLM translate nodes need a paid API call to Zhipu/Z.ai's platform. Privacy is the other reason people pick it: nothing about your prompt text ever leaves your machine.

The trade-off is quality and coverage. Argos Translate's models are smaller, community-trained NMT models rather than a hosted service backed by a huge company's infrastructure, so translation quality on nuanced or slang-heavy text tends to be rougher than Google or an LLM-based translator, and language coverage is narrower - this node's from_translate list tops out at 44 languages against the 100+ on the Google-backed nodes.

How it works

Argos Translate downloads a small neural translation model per language pair the first time you use it, then runs inference on your own CPU/GPU from then on - no further network calls needed for that pair. This is the detail that trips people up on first use: the very first translation for any given language pair will pause to fetch that pair's model, and if you're offline at that exact moment, it'll fail. Warm up your language pairs once while you do have a connection, and it's smooth sailing after that.

The inputs and outputs that matter

• text - your prompt, multiline.
• from_translate (default russian) - source language, spelled out (english, albanian, arabic, …), 44 choices. Note the default here is Russian rather than auto - Argos Translate doesn't do auto-detection the way the Google-backed nodes do, so you need to actually pick your source language.
• to_translate - locked to english (the only choice in this node's schema). If you need a different target language, this specific node isn't the one - check whether the underlying Argos Translate models you've downloaded support other pairs, since this node's fixed choice suggests it's scoped to the everything-to-English use case.

Output is a single text string, same shape as the other translator nodes in this pack.

How to install it

Via ComfyUI Manager: search "ComfyUI Custom Nodes AlekPet" and install. Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/AlekPet/ComfyUI_Custom_Nodes_AlekPet

Restart ComfyUI - the argostranslate package installs automatically on first launch per the pack's install notes. The language models themselves download on first use per language pair, separately from the initial node install - budget a little extra time and bandwidth the very first time you translate from a new source language.

Common issues & troubleshooting

First run of a new language pair hangs or fails. That's the model download happening. If you're offline, or behind a firewall that blocks the download source, it'll error out. Run it once while connected before you rely on it in an air-gapped environment.

Translation reads noticeably worse than Google Translate for the same sentence. Expected - this is a much smaller, community-maintained model, not a byproduct of a search-engine-scale translation system. For anything where nuance matters, GoogleTranslateTextNode or the LLM-backed ChatGLM4TranslateTextNode will generally do better; reach for Argos specifically when offline operation matters more than translation quality.

Can't find your language in the dropdown. 44 source languages is meaningfully smaller than the 100+ on the Google/Deep Translator nodes - if yours isn't listed, those are your fallback.