💬GLM3_turbo_CHAT

This is the least glamorous member of the three-node JcandZero/ComfyUI_GLM4Node pack: a text box that phones home to Zhipu's glm-3-turbo API and drops the model's reply into your graph as a string. That's the whole job. No image input, no system prompt, no local model, no VRAM. You type, it answers, a string comes out the other end.

Honestly, if you want a chat LLM inside ComfyUI you probably want the sibling GLM4_CHAT node instead - it's the same code calling a newer model. glm-3-turbo was Zhipu's fast-and-cheap tier back when this pack was written in early 2024, which makes this node the budget option of a small family. It still works, it's just rarely the one you actually reach for.

How it works

Same machinery as GLM4_CHAT, same file. The node builds a ZhipuAI client from your API key and calls chat.completions.create with the glm-3-turbo model name. One quirk worth knowing: before your actual prompt, the code stuffs in a hardcoded little scripted conversation - "你好 / 我是人工智能助手 / 你叫什么名字 / 我叫chatGLM" - so the model shows up already primed to act like a chat assistant. Harmless, and you can't edit it, because there's no system-prompt input at all.

The inputs that matter

There are only three, all required:

• prompt (multiline STRING) - your instruction. Defaults to "你好，你是谁呀". Use the multiline box to paste whole instructions.
• model_name - a dropdown with exactly one choice, glm-3-turbo. Locked. If Zhipu rotates that model out of its lineup, you can't switch it from the UI; you'd be editing source.
• api_key - your Zhipu key. Defaults to whatever's in the pack's config.json, which ships as a placeholder.

Output: Text (STRING) - the model's reply. Wire it into any text input downstream, like a positive or negative prompt field.

Install

You install the whole pack, not this node alone - all three nodes ship in one repo:

cd ComfyUI/custom_nodes
git clone https://github.com/JcandZero/ComfyUI_GLM4Node
cd ComfyUI_GLM4Node
pip install -r requirements.txt   # just installs `zhipuai`

Or, easier: ComfyUI Manager → Custom Nodes Manager → search "GLM4Node" → install → restart ComfyUI. The node appears in the BlinkNodes_PROMPT menu category.

No model downloads, no VRAM - this is an API node. You need a Zhipu API key and an internet connection. And take the README at its word: the key shown in the author's demo video is dead ("The API I included in the video cannot be used"). Get your own.

Where people get burned

• The model name is frozen to early 2024. The pack hasn't been touched since February 2024. If your call comes back with a model-not-found error, that's the age of the pack, not something you did wrong.
• Your key rides along in the workflow JSON. ComfyUI saves node input values into saved workflows and exported PNG metadata. Don't share a workflow that has your API key typed into the node.
• The error message is a dead end. If the key is missing, the code prints "Error: API key is required" and returns an empty string - no dialog, no hint.

My take: the pack's three nodes share a skeleton, and this one is the skeleton's oldest model. If it works for you, great; if it errors, skip it and use GLM4_CHAT for the same job on a better model.