FLUX.1-Kontext-TextToImage

FLUX.1-Kontext's whole selling point is consistency - it conditions on a reference image so a character or object survives across generations. So the first thing to know about this node is what it doesn't do: there's no image input on it. FluxKontextTextToImage is the plain text-to-image half of the pair, calling Azure's images/generations endpoint with your FLUX.1-Kontext-pro deployment. Same thin HTTP wrapper as the rest of the azure-flux-comfyui pack - prompt in, IMAGE tensor out, no model files, no GPU.

Reach for it when you're already paying for a Kontext-pro deployment on Azure and you want straightforward hosted generation inside a bigger workflow: a draft stage, a batch, a step that feeds something downstream. If you actually want Kontext's reference-image party trick, you want the ImageToImage sibling from the same pack - that's where the model's real capability lives.

How it works

Identical plumbing to the pack's other T2I node. A POST to {endpoint}/openai/deployments/{deployment}/images/generations?api-version={api_version} with your key in an Api-Key header, base64-encoded images come back, and the node decodes them into a batched float IMAGE tensor [N, H, W, C] in 0–1 range. Kontext-pro hosted on Azure is just the model sitting behind that endpoint; the node doesn't know or care.

The inputs that matter

• azure_endpoint - your Azure resource URL. The code adds a missing trailing slash for you.
• api_key - your Azure OpenAI key, stored in plain text in the node.
• deployment - defaults to FLUX.1-Kontext-pro and must match the deployment name in your Azure resource.
• api_version - 2025-04-01-preview by default.
• editing_prompt - yes, the name is recycled from the edit node. Here it's just your text prompt, and the default is a placeholder photorealistic portrait you'll want to change.

Optional: size (1024x1024 format, validated before it's sent), n (1–8, billed per image), and output_format (png/jpeg/webp). The single output is images, wired into Save Image or anything downstream.

Installing it

Shared with the whole pack. ComfyUI Manager → search "azure-flux-comfyui", or:

cd ComfyUI/custom_nodes/
git clone https://github.com/xpanceo-team/azure-flux-comfyui.git
cd azure-flux-comfyui
pip install -r requirements.txt

Restart ComfyUI after. The dependencies are requests, Pillow, torch, numpy - all already in ComfyUI, so nothing heavy downloads. The actual prerequisite, again, is the Azure side: a resource with FLUX.1-Kontext-pro deployed.

Where people get burned

• The 401/403, 404/400, and size-format failures are the same across the pack: wrong key/endpoint, deployment name that doesn't exist, or size not in WIDTHxHEIGHT form.
• Trying to pass a reference image here. It can't take one. If your workflow is feeding an image in and expecting Kontext-style consistency, you've got the wrong node - that input lives on the ImageToImage variant.
• Confusing the hosted pro with the local dev model. The local FLUX.1-Kontext [dev] weights exist, but by mid-2026 the community had moved past it as an editor (Klein and Qwen-Image-Edit took the job), and it's non-commercial licensed and censored anyway. This node calls the hosted pro tier: no weights, filtered at Microsoft's end, billed per image. Same family name, genuinely different thing.

Standard API-wrapper reminder applies: your prompt leaves the machine, and it's metered per call. Keep that in mind before you wire n=8 into a loop.