qwen2 / createImage

The display name says "qwen2," but make no mistake about what this node runs: Qwen-Image, Alibaba's 20B Apache 2.0 image model, reached through fal and billed in Buzz. This is the generation half of the Qwen image family - the big, admired text-to-image model that the community never made its daily driver locally because it's heavy, but that shines at exactly the things 20B parameters are good at: long detailed prompts, natural-language phrasing, and rendering text inside images. Via this node you get all of that without hosting a 40GB+ model.

The node's shape is Qwen's own API shape, and it's refreshingly opinionated. enable_prompt_expansion defaults to True - the API rewrites and enriches your prompt before generation, which is great for vague prompts and mildly annoying when you've written something specific. enable_safety_checker defaults to False, which tells you the API ships with the guard off by default (that's Alibaba's call, and it makes this node notable in the pack). The one control you'll actually think about is image_size, defaulting to square_hd with five other presets - Qwen's sizes are named presets rather than free-form width/height, which is honestly refreshing.

The inputs that matter

• prompt - required. Written in natural language; Qwen-Image rewards descriptive sentences more than tag soup.
• enable_prompt_expansion - on by default. Leave it on for short prompts, turn it off when your prompt is exactly what you want (expansion can flatten a carefully-chosen negative or add flourishes you didn't ask for).
• enable_safety_checker - off by default. Turn it on if you're generating for a platform that needs the belt-and-suspenders.
• image_size - square_hd default, six presets covering portrait/landscape/ultrawide in standard and HD variants.
• negative_prompt - supported; with a 20B model it's a fine-tooth tool rather than a necessity.
• operation - an enum that defaults to createImage (the edit variant lives in the sibling node).
• quantity, seed, output_format, image_metadata - standard.

Outputs: images (IMAGE), errors, workflow_id, raw_json.

Install and auth

The one pack install covers this and the rest: ComfyUI Manager → Civitai Comfy Nodes, or:

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

Authenticate (Civitai account with Buzz): export CIVITAI_API_TOKEN=..., a Civitai Auth node, or the sidebar sign-in.

Where people get burned

The most common surprise is prompt expansion rewriting something you spent time on - when the output comes back looking close but not quite, check whether the expansion toggle is on. Second: the 20B model is slow on the hosted side too; a Qwen-Image call isn't a turbo draft, so budget your Buzz accordingly and don't use it for iteration when a cheaper engine in the pack will do. And a licensing note that's good news for once: the whole Qwen-Image family is Apache 2.0, so using this node doesn't drag you into a closed-source licensing conversation the way some pack neighbors do. If you want the family's real showpiece, the qwen2 / editImage sibling is where Qwen-Image earned its reputation - this node is the capable, well-rounded generator half.