vlo Memory Load Video

Every video-in, video-out workflow in ComfyUI starts the same way: Load Video, file on disk. That's fine for the person dragging a clip into their own install. It's wrong for an app that generates or edits dozens of clips per session and wants each one fed into the graph - because every one of those files becomes a permanent resident of your input folder. vlo Memory Load Video is the no-disk version: the video arrives in an in-memory registry, and this node hands it to your graph as a VIDEO by id.

It's the bridge from vlo, the local open-source video editor this pack serves. vlo is a timeline editor with ComfyUI as its generation engine - you cut, mask, and retime clips in the editor, then push the relevant selection into a workflow. This loader is the receiving end: vlo uploads the clip to the registry over HTTP, the dropdown here lists the ids, and the node decodes the bytes into ComfyUI's VIDEO type (frames plus frame rate and audio, via PyAV) without anything ever landing in input.

Inputs and output

• file - the dropdown of registry ids (videos only). Refresh re-pulls the list from vlo.
• disable_in_memory (boolean, default false) - flip it and the dropdown lists files from your normal ComfyUI/input folder instead. That's your hand-testing path when vlo isn't running.

One output: VIDEO - the standard ComfyUI video object, which carries its frame rate and any embedded audio, ready for vlo Video Convert FPS, samplers, or the websocket savers downstream.

What bites people

The registry is shared with the other memory loaders, and its rules are the things that bite: 512 MB per item, 2 GB total, unread items expire after two hours, accessed items after ten minutes, and the whole thing is wiped when ComfyUI restarts. A run that fails with "Unknown media id" almost always means the item aged out or the server restarted - re-send the clip from vlo. And since video is the heavy one, it's the most likely to trip the per-item size cap on long or 4K clips; if you're hitting 413 errors, that's the cap talking.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/PxTicks/ComfyUI-vlo.git

Restart ComfyUI (or use ComfyUI Manager, searching "ComfyUI-vlo"). The web extension registers itself. No model downloads; the video decoding stack ships with current ComfyUI, so nothing extra to pip install.