AI_Qwen

This is a vision-language chat node against Alibaba's Qwen-VL models, called directly through Alibaba Cloud's DashScope API rather than through ModelScope's hub (that's what separates it from AI_ModelScope_image in the same pack - same broader Alibaba/Tongyi family, different API surface and a different account you'd sign up for). Hand it up to three images and a question, get a text answer back - description, analysis, comparison, whatever the prompt asks for.

Qwen is one of the two or three labs that made 2025 the year Chinese open-weight models became the community's default (Wan for video, Z-Image for photorealism, Qwen-Image for editing, all Apache 2.0 and largely uncensored, filling the gap Western labs left as they locked things down) - and DashScope is Alibaba's commercial API surface for the closed/hosted end of that same model family, including vision variants that aren't necessarily released as open weights at all.

Required fields: model (default qwen-vl-max-latest, with qwen3.5-plus, qwen3.6-plus, and qwen-vl-max as alternates - note that -latest suffix means the exact model version behind it can shift over time without you changing anything, which is convenient until it isn't), preset (defaults to "None" until you've saved one via AI_PresetSave), custom_system_prompt, and text, your prompt. Optional: image_1/image_2/image_3, max_tokens (10–8192, default 1024), seed, api_key_input, and custom_model for a free-text override. Outputs: result and system_prompt.

Installing it: through ComfyUI Manager (search "ComfyUI-Apt_Preset") or manually - cd ComfyUI/custom_nodes && git clone https://github.com/cardenluo/ComfyUI-Apt_Preset - then install.bat and restart. Pure API client, nothing to download locally.

Getting a key: sign up for Alibaba Cloud and enable DashScope, then generate an API key from the console. This is a different signup path from ModelScope's - don't assume a ModelScope token works here, or vice versa; the two nodes in this pack hit genuinely separate Alibaba services even though they're both ultimately Qwen models under the hood.

Troubleshooting: an auth failure means check api_key_input first - confirm it's a DashScope key specifically, not a ModelScope token, since those are easy to mix up if you're juggling several of this pack's API nodes at once. If you picked qwen-vl-max-latest and results seem inconsistent between sessions weeks apart, that's the floating -latest alias doing exactly what it's meant to - pin to qwen-vl-max (without the suffix) if you need a more stable target for reproducible testing. And if the model seems to be ignoring an image, double-check it's actually connected to image_1 and not a later slot; the schema treats these as ordered inputs, not an unordered pool.