qwen / createImage

Qwen-Image is the model everyone respects and almost nobody runs as a daily driver - Alibaba's 20B diffusion transformer with the best-in-class text rendering, so heavy that most people keep it out of their local pipeline. This node gives you the hosted version on Civitai's orchestration layer: the typography capability ("commercial-grade text in both English and Chinese" is the KB's summary), the strong prompt adherence, none of the VRAM. Billed in Buzz.

There's a twist, though. This is hosted Qwen-Image, and the model selector defaults to 2.0 - with 3.0-pro, 2.0-pro, and max/plus options. Qwen-Image 2.0 is the API-and-chat-only successor that never shipped open weights, so this node is actually one of the few ways to reach the newer closed Qwen line rather than the open 20B checkpoints you could run locally. Keep that in mind: model: 2.0 is not the open Qwen-Image you may have read about.

How it works

Standard civitai-comfy-nodes flow: submit the prompt (plus width/height) to the Civitai Orchestration API as a qwen / createImage job, long-poll, return an IMAGE tensor. The notable defaults: prompt_extend is on by default (the API rewrites/expands your prompt before generating) and watermark is off. If you want literal control over what's generated, prompt_extend is the first switch to throw.

The inputs that matter

• prompt, width, height - all three required. Note that width/height default to 0 (meaning "API default" - Qwen's API handles resolution internally); set them explicitly only if you need a specific size.
• model - 3.0-pro, 2.0-pro, 2.0 (default), max, plus. The newer numbers are the closed line.
• prompt_extend - boolean, default True. Turn off for literal prompts.
• watermark - boolean, default False.
• negative_prompt - supported here (Qwen's API has one).
• quantity (1–6), seed, output_format, image_metadata - standard.

Outputs

images → Preview/Save Image. errors, workflow_id, raw_json are diagnostics.

Installing and authenticating

Pack-wide install:

cd ComfyUI/custom_nodes
git clone https://github.com/civitai/civitai-comfy-nodes.git
pip install -r civitai-comfy-nodes/requirements.txt

or ComfyUI Manager → "Civitai Comfy Nodes" → Install → restart. Dependencies: requests + python-socketio[client].

Auth: Civitai Auth node → CIVITAI_API_TOKEN → stored key/OAuth → browser login. Headless:

export CIVITAI_API_TOKEN=your_token

Where people get burned

• prompt_extend on by default. People paste a precise prompt, get an embellished result, and blame the node. It's the checkbox. Turn it off for exact wording.
• Width/height at 0 is intentional. The API picks resolution; fighting it with explicit pixels can produce odd results on a hosted pipeline that re-sizes internally.
• The default model is the closed one. If you wanted open Qwen-Image weights, this node is the wrong door - and max/plus are a different (and pricier) tier of the hosted line. Check which model you're on before assuming output behavior.
• Billed per image. quantity up to 6 multiplies the Buzz.
• Your prompt goes to Alibaba's API via Civitai's orchestration. Standard API-node deal - and hosted moderation applies.