Qwen2-VL Text Preview

This is the smallest node in the pack and it exists to solve a genuinely annoying ComfyUI quirk: STRING outputs are invisible by default. Run one of the two "QWEN VLM MACHINE" nodes and you'll get your caption or description back as a text output - and then see absolutely nothing on screen. The machines aren't output nodes, so ComfyUI has no reason to draw the text, and a fresh workflow can make the whole thing look broken. Qwen2-VL Text Preview is the missing display: it takes a string, shows it in the UI, and passes the same string through so the graph keeps flowing.

It's an output node in the plumbing sense - the thing that marks a branch as "show me the result here" - and it does exactly one job. You wire the machine node's text into this node's text input, run the queue, and the description appears in a little panel.

Why it's useful beyond this pack

Don't overthink this one. It's a generic string viewer - the input is a plain STRING with force-input enabled, so it isn't tied to Qwen2-VL at all. Any node in any pack that emits text can drop into it, which makes it a handy general-purpose display to have around even if you never load a vision model. The output is a passthrough of the same string, so you can chain it mid-graph and still wire the value onward into, say, a text encode node.

The one real gotcha: it's a display node, not a storage node. It shows text in the UI but doesn't save it anywhere. If you want the caption written to disk, wire the string into a text-save node (or feed it into the filename machinery) on a separate branch. The preview is for seeing what the model said, not for keeping it.

Installing

It ships in the same pack as the machines:

cd ComfyUI/custom_nodes
git clone https://github.com/darkpool999/ComfyUI-Qwen2-VL-Nodes
cd ComfyUI-Qwen2-VL-Nodes
pip install -r requirements.txt

Restart ComfyUI and it appears under the Qwen2-VL category, or install via ComfyUI Manager by searching "Qwen2-VL". You don't need llama-cpp-python for this node - the base requirements are plenty.

If you wire everything up, queue, and still see no text, check the obvious first: is the machine node actually running? These packs download a multi-gigabyte model on first run, so the first execution can sit there for a while before any text exists to preview. Once it produces, the panel fills in.