Qwen2.5-VL-7B Age Detector

This is the node you don't build a workflow around - you bolt it on as a bouncer. Feed it any image and it looks at the face and tells you, in plain text, how old the person probably is. It's the Runware-hosted version of Alibaba's Qwen2.5-VL-7B, a small vision-language model, pointed at the specific job of age estimation.

Why does that exist inside a ComfyUI pack at all? Content safety, mostly. If you're auto-generating or auto-sharing images, or cleaning up a dataset, you want a cheap filter that flags minors before an image goes anywhere. The pack already surfaces NSFW flags on the node title bar; this is the age half of that story. It's also handy for dataset tagging - caption thousands of face crops and get an age label for each.

How it works

Under the hood there's not much to it. Your IMAGE tensor gets encoded to a PNG and shipped to Runware's API as a caption task (runware:152@50), the hosted Qwen2.5-VL-7B looks at it, and the response comes back as a string on the text output. The whole thing runs in the cloud, so the only hardware cost is on Runware's side - your GPU does nothing.

There's exactly one input that matters:

• image - the IMAGE tensor to inspect. Wire in a Load Image, or anything upstream producing an IMAGE.

And one output:

• text - the age assessment as a STRING. Read it on the node, or wire it into a conditional/display node.

Installing

It's the same install as every other node in the Runware pack - there's one repo and about 350 nodes in it. ComfyUI Manager → Custom Nodes Manager → search Runware → install → restart. Or manually:

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

Dependencies are light (the runware-sdk, Pillow, soundfile) and there are no model downloads - the model lives on Runware's servers. You do need an API key from the Runware dashboard, set via ComfyUI Settings → Runware API key, the RUNWARE_API_KEY env var, or runware auth login.

Where people get burned

Three things. First, the output is a free-text string, not a number - the model answers in natural language like "approximately 25-30 years old". Don't feed it straight into anything that expects a float. Second, it's an estimate. A 7B vision model guessing ages from faces will confidently be wrong on kids with adult styling, heavy makeup, and anyone whose face doesn't match the training distribution. Use it as a moderation triage tool, not a legal one. Third, it's a paid API call per image - cheap, but if you batch a thousand crops it adds up, and the node's title bar shows you the cost per run so you can watch it.

If you only need this occasionally, it's genuinely the easiest age detector you'll ever wire up - no model file, no ONNX, no VRAM. If you need it at scale and free, that's a different conversation; this is the "works in five minutes" option.