Memories Video Age Detection

Some workflows need to know things about a video that pixels alone don't make obvious - like roughly how old the person in frame is. This node takes a video, sends it to a "memories" vision model on Runware's cloud, and returns its age estimate as plain text. It's niche, but when you need it - content moderation, dataset curation, a sanity check before you process a batch of face videos - nothing else in the pack does it.

How it works

This is a caption task in the Runware pack, which means it goes through the same base machinery as the image caption nodes: the request hits Runware's inference API, the model watches the video, and the answer comes back as a text output (STRING). You wire that into a ShowText-style node or any text consumer.

The single input is video, and it's a string - a UUID or URL, not a VIDEO tensor. That's the pack's convention for media inputs that aren't generated in-graph: the README spells it out (audio, video, and document inputs take a URL, path, or UUID). So if your video lives on disk, you'll need it hosted somewhere reachable, or uploaded to your Runware account first, before you can point this node at it.

The output is text, so don't expect a structured breakdown. You get whatever the model returns as its answer string; if you need the estimate machine-readable, you're on your own parsing it - a lightweight node of your own can do the job.

Install

This is one node in ComfyUI-Runware, so install the pack once:

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

Restart ComfyUI, or use ComfyUI Manager → search "Runware" → install. Nothing heavy downloads - the vision model runs in the cloud, so any machine works. You do need a Runware API key: paste it in Settings → Runware API key, set RUNWARE_API_KEY, or run runware auth login once.

Gotchas

First, it's pay-per-run and network-dependent - the title bar shows the cost after each call. Second, and this is the one that trips people up: video inputs are strings, not VIDEO tensors. If you're coming from a workflow where a video node produced a VIDEO output, you can't wire that straight in; you need a URL or a Runware-hosted UUID. And finally, "age detection" here means the model's best guess from appearance - treat it as an estimate, not ground truth, especially if you're using it for any kind of compliance decision.