CXH_DeepLX_translate

This is the sibling that actually has a setup step. The pack's README is two lines long, and the gist is: go to https://deeplx.missuo.ru/, get a token, paste it into key.txt in the node folder. Once that's done, CXH_DeepLX_translate gives you what the free node can't - four target languages, source auto-detection, and a translation path you actually control.

The mechanism: where CXH_DeepLX_Free calls DeepL's web endpoint directly, this one talks to a DeepLX server. DeepLX is a well-known project that reimplements DeepL's API so you can self-host a translation endpoint; missuo.ru runs one publicly, and this node POSTs to it. Each run it opens key.txt in the pack directory, appends the key to the proxy URL, and sends:

{"text": "...", "source_lang": "...", "target_lang": "..."}

Then it pulls the data field out of the JSON response and returns it as a STRING. The node reads key.txt fresh on every execution - and prints the key to the console while it's at it. Minor annoyance, and a real privacy note: your prompts are going to a third-party server.

The inputs:

• text (multiline) - what you're translating
• source - auto, ZH中文, EN英语, DE德语, JA日语; default auto
• target - ZH/EN/DE/JA; default EN英语
• proxy - default https://deeplx.missuo.ru/translate?key=

That proxy field is the quiet star. Defaulted, it points at missuo's public server; you can repoint it at your own self-hosted DeepLX (OwO-Network/DeepLX) and skip the third party entirely. If you're serious about this, that's the setup: run DeepLX on your machine or a VPS, point the node at it, done. Public DeepLX servers come and go, get slow, and go down - and sending every prompt you translate through a stranger's box is a gamble. ComfyUI already runs arbitrary custom-node code on your machine with no sandbox; don't add a data-exfil channel on top of that for no reason.

The output is one STRING named out, which wires into the same places as the free node: CLIP Text Encode, a text display, anything that takes text. That makes it handy for the same Chinese↔English prompt swapping - models with Chinese-native encoders respond to a Chinese prompt very differently than the same English one - just with more languages on the menu.

Troubleshooting worth knowing: this node fails soft. If the server returns anything but 200, it returns the HTTP status code as your "translated" string; on an exception it returns the literal string "error". Your workflow won't crash - it'll just silently feed "404" into your prompt encoder. If your renders suddenly contain status codes where prose should be, check the token or the server first. Also, target values like "EN英语" are truncated to their first two letters internally, so don't try to sneak custom language codes past the dropdown.

Install is identical to its sibling. ComfyUI Manager, search "Comfyui_CXH_DeepLX", or:

cd ComfyUI/custom_nodes
git clone https://github.com/StartHua/Comfyui_CXH_DeepLX

then restart and fill key.txt - it ships empty. The requirements.txt lists PyDeepLX, but this particular node actually just uses requests, which already ships with ComfyUI; that dependency line exists for the free node. No models, no weights, nothing heavy.

The verdict: if the free node's ZH/EN-only limit is too tight, or you want to self-host your translation, this is the node. Just treat the default proxy as a demo endpoint, not infrastructure.