Azure OpenAI GPT (Factx API)

FactxAzureOpenaiGPT (2lab) is the Azure-flavored member of the pack's Factx API family. Same shape as the other Factx LLM nodes - credentials in, prompt in, generated text out - except the model behind the scenes is served through Microsoft Azure's OpenAI offering, reached via the Factx marketplace rather than an Azure subscription you own.

If you're not already deep in the China-facing API ecosystem, the thing to understand is what makes this node different from a plain OpenAI call. Azure OpenAI doesn't expose models the way the standard API does; you create deployments - named instances of a model - and you call those by name. That's exactly why this node has the deployment field that the other Factx LLM nodes don't. It's not a model picker; it's the name of a deployment that exists in the Azure account behind your Factx credentials. The author's use case is presumably people whose API access to Azure OpenAI is brokered through Factx, so the deployment name refers to the setup on the provider's side.

How it works

Required inputs:

• api_id - your Factx account ID, a masked KEY field.
• api_key - your Factx API key, also masked.
• prompt - the message, multiline.
• deployment - a KEY-typed field for the Azure deployment name. This is the field to get right: if it doesn't match a deployment available through your Factx credentials, the call fails with a "deployment not found"-style error, not a useful "you typed the model wrong."

The text output is the model's reply, ready to wire into your prompt pipeline or wherever the graph consumes text.

Honest expectations, same as the rest of the Factx family: the pack is undocumented (README one word long), there's no meaningful English community footprint to lean on for troubleshooting, and the exact Azure model behind your deployment is defined by whoever configured it - not by anything in this node. If you hold your own Azure OpenAI keys, you'd typically be better served by a node that talks to Azure directly; this one only makes sense as part of a Factx-brokered setup.

On the plus side it's still just an HTTP call - no local weights, no VRAM, no download. And the key hygiene note applies to every node in this family: masked KEY fields keep keys off the visible graph, but keys stored in workflow JSON still get embedded into output images. Keep live keys out of workflows you share.

Installing it

From 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 after. No models to download, no dependency fights. If your deployment name matches what your credentials actually expose, this is a working Azure-OpenAI-in-the-graph node; if it doesn't, the error will point at the field to fix.