Argos Translate CLIP Text Encode Node

This is ArgosTranslateTextNode - the fully offline, no-API-key translator built on the Argos Translate library - fused with a CLIP Text Encode. Feed it a non-English prompt and a CLIP model, and it hands back conditioning without ever making a network call, because the translation runs entirely on your own machine once the language model for that pair is downloaded.

Why you'd reach for it

Same case as the standalone Argos node: you're on an air-gapped or locked-down machine, or you'd rather your prompt text never leave your box, and you don't want the extra node clutter of a separate translate-then-encode pair. It's the privacy/offline option among this pack's three CLIP-encode translator variants (Google, Deep Translator, Argos) - trading translation polish for zero network dependency once warmed up.

Go in with the same expectation as the plain text version: Argos Translate's community-trained NMT models are smaller and rougher around slang and nuance than Google's or an LLM's, and this node's language coverage tops out at 44 source languages, all mapping to English - narrower than the Google or Deep Translator equivalents.

How it works

On first use for a given language pair, Argos Translate downloads a small dedicated translation model; after that, translation runs locally with no further network calls. That translated text then flows straight into a standard CLIP encode against the CLIP model you provide.

The inputs and outputs that matter

• text - your prompt, multiline.
• clip (CLIP, required) - from your checkpoint loader.
• from_translate (default russian) - source language, 44 choices, spelled out in full (english, albanian, arabic, …). No auto option here - you need to pick the source language explicitly.
• to_translate - locked to english, the only choice this node's schema exposes.
• Outputs: CONDITIONING (into your sampler) and STRING (the translated text - plug into a PreviewTextNode to sanity-check it, especially useful here since Argos's quality varies more by language pair than the API-backed translators).

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. Language models for whichever pairs you actually use download separately, on first use of that pair - do this once while you have a network connection if you're planning to run fully offline afterward.

Common issues & troubleshooting

First run for a language you haven't used before hangs or fails. That's the language-pair model trying to download. If you're offline at that moment, or a firewall blocks it, expect an error. Run each language pair you need at least once while connected before relying on this fully offline.

Encoded output looks off compared to what you'd get through Google. Argos's models are meaningfully smaller and less polished than Google's or an LLM's - that's the trade for offline operation, not a bug. If translation quality on a specific phrase matters more than staying offline, swap to GoogleTranslateCLIPTextEncodeNode or ChatGLM4TranslateCLIPTextEncodeNode for that run.

Your source language isn't in the dropdown. 44 languages is the ceiling here; the Google- and Deep-Translator-backed variants both cover 100+ and are the fallback if yours is missing.