CXH_DeepLX_Free

The name is half a lie, and that's the good half. CXH_DeepLX_Free doesn't call any paid API and needs no key, token, or account. "Free" here means the node talks to DeepL's free web translator directly through the PyDeepLX library - the same reverse-engineered endpoint the browser version of DeepL uses. You feed it a string, it hands you a string back, done.

Why would you want that inside ComfyUI? Prompt translation. The node is a wire-in-your-graph translator for English and Chinese. If you've ever run a workflow around a Chinese-community model - Kolors, Hunyuan, anything with a Qwen-based encoder - you know those text encoders read a prompt in Chinese very differently than the same prompt in English. People translate prompts to Chinese for exactly that reason; on LLM-encoded models your prompt is read as an instruction, not a tag bag, and which language it's in genuinely changes the output. This node lets you do that on the fly, or keep one workflow that flips the whole prompt to English for a different model, without copy-pasting into a browser tab.

How it works is almost too simple to explain. The node calls PyDeepLX.translate(text, source, target) and returns whatever comes back. The only flourish is a try/except that just retries once on failure. That's the entire node. It's a wrapper around a wrapper - PyDeepLX wraps DeepLX, which reverse-engineers DeepL's free web API - and the trade-off is baked in: free means rate-limited, and whenever DeepL shuffles its web endpoint (it does, periodically), PyDeepLX needs a bump before the node works again. If it stops translating one day, check for a PyDeepLX update before you suspect your graph.

The inputs are the short list:

• text (multiline) - the string to translate
• source and target - each limited to ZH or EN, default ZH → EN

That's the real limitation, honestly: two languages only. Need German, Japanese, or auto-detection? The sibling CXH_DeepLX_translate node in the same pack covers those, at the cost of needing a token.

The output is a single STRING named out. Wire it into a CLIP Text Encode positive prompt, drop it into a text display node, or chain it further down the graph. That's the whole loop.

Install is standard. ComfyUI Manager → search "Comfyui_CXH_DeepLX" → install, or:

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

then restart ComfyUI. Manager installs the single dependency, PyDeepLX, for you. No model files, nothing heavy to download. And for this node you never touch the key.txt the repo ships with - that file belongs to the translate sibling. Don't confuse the two: one needs nothing, the other needs a token.

Realistic verdict: for a zero-setup Chinese↔English prompt swap this is the one you'd actually reach for. Just don't build anything production-critical on it - a reverse-engineered endpoint can vanish overnight.