Azure OpenAI GPT

Two nodes in this pack share the display name "Azure OpenAI GPT," and they could not be more different under the hood. The one that looks complete - Azure_Openai_GPT (2lab) - takes key, endpoint, deployment, and version, and calls Azure with your credentials. This one, AzureOpenaiGPT (2lab), takes just a prompt and a deployment name. No key. No endpoint. No version. Which means it's the server-backed variant: your prompt goes to the 2lab cloud backend, and the backend - which already holds Azure credentials - makes the call with the deployment you named.

The inputs and outputs that matter:

• prompt - the text you're sending, multiline.
• deployment (KEY) - the Azure deployment name the server should hit. It's a KEY field, so it's entered as a hidden value.
• text out (STRING) - the model's reply, for your prompt chain.

Two inputs, one output. That's the whole node.

The pattern should look familiar by now if you've been reading about this pack: every "api" node without a key field is really a 2lab-cloud node. The trade is the same as with ChatGLM (2lab) - you get to use the service without managing credentials locally, but you're tied to the 2lab platform, which means the WeChat QR binding (2lab_key.png) on first launch, a Chinese-market backend, and effectively zero English documentation or community if something goes wrong.

My honest advice: if you have your own Azure OpenAI account, you don't want this node - use Azure_Openai_GPT (2lab) instead, where you control key, endpoint, and api-version directly and can debug failures yourself. This variant makes sense only if you're building workflows to publish on the 2lab platform and you want the app's users to trigger Azure calls without ever seeing an Azure credential. For the standalone ComfyUI user, it's a node that will sit in the menu and stare at you, useless without the 2lab account.

A note on the naming, because it will bite you: Azure_Openai_GPT and AzureOpenaiGPT render nearly identically in search and in menus. When a downloaded workflow says "missing node AzureOpenaiGPT (2lab)," make sure you install the pack and identify which variant the workflow actually used - the fully-keyed one will ask you for credentials, the server-backed one won't.

Installing it

Part of AI2lab/comfyUI-tool-2lab. ComfyUI Manager → search "comfyUI-tool-2lab" → install → restart, or:

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

Restart ComfyUI. No model downloads; the compute happens server-side at the 2lab backend.

Common issues

If you see an error about a missing user key, you're not bound to the 2lab platform - that's the WeChat step you skipped, and this node can't work without it. If you're bound and the call fails, the deployment name you typed has to match one the backend's Azure account actually owns; a wrong name gives a server-side error that the node may pass through as raw JSON. Debugging depth ends there - the pack is a Chinese-market tool with a thin README and no English community writing about it, so this node is really only for people already committed to the 2lab ecosystem.