Azure OpenAI GPT

If you're running LLM calls through Microsoft Azure's OpenAI service - common when your employer's compliance team says no to every other option - this node is how you thread it into a ComfyUI graph. It takes the five things Azure OpenAI always demands, sends your prompt, and returns the model's text as a STRING for your workflow.

Azure OpenAI is the corporate version of ChatGPT-as-an-API: instead of one global endpoint, every model lives behind your own resource's URL, keyed by a deployment name you created, and versioned with an api-version string that the API checks on every call. That's why this node has five required fields where the plain-OpenAI nodes have two. Each one is load-bearing.

The inputs that matter:

• key - your Azure OpenAI API key (hidden KEY field).
• endpoint - your resource's base URL, the https://your-resource.openai.azure.com/ bit. Get this wrong and nothing else matters.
• deployment - the deployment name you created in Azure, which is not necessarily the model name. People paste gpt-4o here and get 404s because their deployment is called gpt4o-prod.
• version - the api-version string, like 2024-02-15-preview. Azure is ruthless about this; a stale or malformed version string is one of the most common silent-ish failures.
• prompt - the text you're sending.
• text out (STRING) - the model's response, wired into the rest of your graph.

The KEY type on the config fields is deliberate: they're entered as hidden inputs, so a screenshot or a shared workflow won't broadcast your credentials to the world. Do still be careful about sharing raw workflow JSON, though - hidden fields can survive in the file.

The honest take: this is the most complete of the pack's LLM nodes, and it's the one you'd use if you have real Azure credentials. If you don't have Azure, skip it. And note the pack also ships AzureOpenaiGPT (2lab) - same display name, different schema, only prompt and deployment - which is the server-backed variant that routes through the 2lab cloud. They're easy to confuse in a menu; this one, with the full field set, is the one that works with your own account.

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 - this is a pure API call.

Common issues

The classic Azure stack: 401 means the key is wrong or the resource doesn't have that model enabled; 404 on a request to your endpoint usually means the deployment name doesn't match what you actually created; 400 with an obscure code usually means the api-version string is stale or invalid for the model. The pack's documentation is a thin Chinese README with no English community to lean on, so when Azure's error messages get cryptic, your best tool is Azure's own docs for the version string and deployment setup - the node itself is a thin, honest wrapper.