Qwen Chat

This is the pack's gateway to Alibaba's Qwen models - the same family behind DashScope, Alibaba's cloud AI platform. Qwen Chat is a chat node: you give it a system instruction and a user prompt, it calls the Qwen API, and you get the reply as text you can pipe anywhere in your workflow. Where it gets more interesting than your average LLM node is the multimodal bit: it takes images, up to ten of them, and turns them into a conversation about what it sees.

That's the genuinely useful mode for a ComfyUI context. Wire a generated image into image_1 and ask "what's the mood here, and how would you describe the lighting?" and the reply flows into a prompt builder, a metadata field, or whatever you're automating. It's vision-to-text inside the graph, no file juggling.

How it works

Under the hood it calls DashScope's OpenAI-compatible chat endpoint (dashscope.aliyuncs.com/compatible-mode/v1/chat/completions) - so if you've used an OpenAI-style API, this is the same shape. Images are converted to base64 data URLs and sent inline as image_url content parts alongside your text. Responses come back as plain text plus the raw response payload.

The fields you'll actually touch:

• api_key - your DashScope API key (from Alibaba Cloud's DashScope console). The node refuses to run without it.
• user_input - your prompt/question.
• system_instruction - optional; use it to set the persona or rules ("you are a prompt engineer who writes detailed tags").
• temperature (0–2) and top_p (0–1) - the usual sampling dials.
• model - defaults to qwen3.7-plus, but that default is worth a skeptical look; check the current DashScope model list because naming there moves.

Outputs are text (the reply) and raw_response (the full JSON - handy when you want to debug or extract metadata yourself).

Install

Ships in ComfyUI-WBLESS:

cd ComfyUI/custom_nodes
git clone https://github.com/LaoMaoBoss/ComfyUI-WBLESS

restart ComfyUI, or install "ComfyUI-WBLESS" via ComfyUI Manager. No models, no local weights - everything runs on Alibaba's servers, so you need internet and a paid key.

Troubleshooting

The number-one stumble is the default model name. qwen3.7-plus is what the pack ships with, and it may or may not match a live DashScope model ID on your account - if you get model-not-found errors, go pull the actual current name from DashScope's model list. A 401 is a key problem (wrong key, or the key doesn't have the model's access enabled - DashScope gates some models behind an application process). And if images aren't being "seen," double-check you're connecting to image_1 and not just typing about the image in the text field. Once it's dialed in, it's the pack's answer to "I want an LLM with eyes, wired into my graph."