ZhiPu Service Connector 🦉| OwlV

ZhiPu (智谱AI) is the Chinese lab behind the GLM family of models, and its API lives at open.bigmodel.cn. The ZhiPu Service Connector is the dedicated path to GLM for the ComfyUI-Translator pack's LLM Translator. Why bother with a dedicated node when the pack's General connector exists? Because the dropdown only lists GLM models, so you can't fumble into asking for a model this provider doesn't have. That's the whole value of the dedicated connectors, and it's real.

Why you'd reach for it

You have a ZhiPu account, or you want GLM specifically. GLM models are strong at Chinese-to-other-language translation and cheap at the token level, and the API works fine on a mainland network with no proxy - the README's domestic path for China users. If you don't care about the provider and just want the best model for the job, the pack's General connector or SiliconFlow option gives you more choices; this node is for when you know you're a ZhiPu shop.

How it works

The node wraps a ZhiPuConnector pointed at https://open.bigmodel.cn/api/paas/v4/chat/completions - standard Bearer auth with your key, payload of max_tokens 512, temperature 0.7, frequency_penalty 0.5. Same pack pattern as all the others: the connector node itself makes no network call. It emits an llm handle, and the actual request fires when the LLM Translator node runs downstream. Run the connector alone and you'll see an object, not text - that's expected.

Inputs and output

• api_key - password-masked in the UI.
• model_select - glm-4, glm-4v, glm-3-turbo, or Custom. Default is glm-4.
• custom_model_id - the model name when you pick Custom.

One interesting wrinkle: glm-4v is ZhiPu's vision model. The LLM Translator only ever sends text, so in this workflow it just behaves like a normal text model - the vision support is there, it's just unused.

Output is a single llm (type LLM) that plugs into the LLM Translator node.

Installing it

ComfyUI Manager: search ComfyUI-Translator, install, restart. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/OwlvChirotha/ComfyUI-Translator
pip install -r ComfyUI-Translator/requirements.txt

Then restart. No model downloads, no heavy dependencies - just requests for this node. The googletrans line in requirements.txt belongs to the pack's Basic Translator.

Common gotchas

• The listed models are stale. glm-3-turbo and glm-4 were current when the author pinned them, but ZhiPu has moved on - there's a whole line of newer GLMs (glm-4-flash, glm-4.5, glm-4.6). Expect to land on Custom with a current model ID from the ZhiPu console rather than living in the dropdown.
• 512-token output cap. Long text gets truncated. Keep translation chunks reasonably sized.
• Errors come back as text. A bad key or retired model shows up as an error string in the LLM Translator's output rather than a crash.
• Your key is in the workflow JSON. The UI masks it, the saved file doesn't. The README says to clear keys before sharing workflows; it means it.

This pack is a one-commit GPL-3.0 hobby project with no community footprint to speak of, and the ZhiPu node is one of its more formulaic connectors - which is fine. If you already have a ZhiPu key and GLM is your model of choice, it's the straightforward way in.