Seed-Nodes: QwenVLAPI

Every so often you want a node that can look at an image and tell you - or a downstream node - what's in it. That's what QwenVLAPI does: it sends your image to Alibaba's Qwen-VL vision model over the DashScope API and returns the model's text answer as a string. Auto-captioning your dataset, verifying an image matches its prompt, asking questions about a render, wiring a "describe this and feed it back" loop - that's this node's territory.

How it works

It's a thin API client. Your image gets converted to a base64 PNG, bundled with your prompt, and POSTed to DashScope's multimodal-generation endpoint (dashscope.aliyuncs.com). The model's text reply comes back as the response string output. Straightforward, and worth knowing three honest limitations:

• It calls the cloud, so it needs a DashScope API key and network access. There's no local model here - you're renting Qwen-VL per request.
• It uses only the first frame of your image batch. No fancy per-frame captioning.
• If the API key field is empty, it doesn't throw an error - it returns a Chinese error string as the "response." Confusing the first time you see it.

The inputs that matter

• api_key - your DashScope key. Get one from Alibaba Cloud's model studio (DashScope); this is the one input you can't skip.
• prompt - the question or instruction. Default is "请描述这张图片的内容" ("describe the content of this image"), but you'll want your own, like "tag this image with booru-style tags" or "does this match the prompt: X".
• model - pick from the Qwen-VL lineup: qwen-vl-max variants for the top-end quality, qwen-vl-plus (the default) for a cheaper, faster balance. The list includes dated snapshots; -latest follows updates.

Optional image is the thing being described. Output is a single response string - wire it into any node that takes text (a text display, a file writer, a prompt builder, a LLM node).

How to install it

Standard pack install - it lives in ComfyUI-Seed-Nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/Aerse/ComfyUI-Seed-Nodes
# restart ComfyUI

Or ComfyUI Manager → search "ComfyUI-Seed-Nodes". No model downloads, no GPU requirement - the model runs on Alibaba's side. You do need that API key, and Qwen-VL is a paid API, so budget accordingly.

Common issues

• Response is a Chinese error message - either the API key is missing/invalid or the request failed; the node routes errors into the output instead of raising them. Read it, then check the key.
• No image attached - the node works without an image (pure text prompt to a vision model), but that's usually a mistake in wiring.
• Cost surprises - high-res images and -max models cost more per call. The qwen-vl-plus default exists for a reason.

If you're already in the Alibaba/Qwen ecosystem, this is a quick, keyed win. If you'd rather keep everything local, it's not the node for you - but for captioning chores it beats running a VL model on your own GPU.