ChatGLM chatGPT 智谱AI

Here's the thing that confuses everyone the first time they meet this pack: there are two "ChatGLM" nodes, and this one has no API key field. Just a prompt in, text out. No key, no model dropdown, no configuration at all. That's because this version doesn't call Zhipu directly - it sends your prompt to the 2lab backend (the pack's cloud service, fronted by the Factx API) and the server does the ChatGLM call on your behalf.

So the mechanism is: your prompt goes over the network to api.factx.cn, the server runs the LLM using its own account, and hands back the text. The cost is paid not by you per-token, but by the fact that you need to be logged into the 2lab platform - bound via that WeChat QR (2lab_key.png) the pack generates on first launch. No binding, no service, and the node raises an error about a missing user key.

The inputs and outputs that matter:

• prompt - the only real input, multiline. This is everything you're asking the model.
• text out (STRING) - the reply, wired into your prompt chain or into a display node.

That's it. It's the "API access for free, at the cost of being in the 2lab ecosystem" version of the pack's ChatGLM node.

Now the honest assessment. If you're a typical reader - outside China, not signed up for the 2lab WeChat mini-program platform, not planning to publish workflows to it - this node is dead on arrival. The binding step requires WeChat and a Chinese-market service, and the backend is aimed squarely at that audience. There's no workaround where you inject your own key, because there's no key field. Your move is to reach for the sibling node, ChatGLM_GPT (2lab), which takes your own api_key and calls Zhipu directly. Same family, same LLM, but it works without the 2lab account.

The only real reason to use this one: you're actively building on the 2lab platform, you've done the WeChat binding, and you want published workflows whose users can call the LLM without the app exposing or requiring an API key. In that context, the server-backed node is actually the elegant choice - the end user's prompt rides on your backend account.

Installing it

Part of AI2lab/comfyUI-tool-2lab. Install via ComfyUI Manager (search "comfyUI-tool-2lab") or:

cd ComfyUI/custom_nodes
git clone https://github.com/AI2lab/comfyUI-tool-2lab

Restart ComfyUI. Requirements are light; there's no local model - this one lives on the 2lab server.

Common issues

The failure mode is unmissable: no bound user key means the node throws something like "还没设置userKey" (user key not set). That's your cue that the WeChat binding step was skipped, not that the node is broken. If you're getting that and you don't intend to use the 2lab platform, stop here and use ChatGLM_GPT (2lab) with your own key. If you are bound and still get failures, it's a network or backend issue - and given the pack has no real English documentation or community, that's about as far as the help goes.