Load Video (Upload) (VideoChat)

VCW_LoadVideo is the front door of the video-chat half of lebrosoft/ComfyUI-VideoChatWrapper, and the first thing to know is that the name oversells it. It doesn't load a video into the graph. It doesn't decode frames, extract audio, or do anything clever. It's a file picker that scans your ComfyUI input folder, lets you choose one of your videos from a dropdown, and hands you the file path as a plain string. That's the whole job, and in the context of this pack it's the correct job - the summary node downstream needs a path it can hand to a vision model, not pixels.

How it works

On startup, the node lists ComfyUI/input/ and filters for four video extensions: .webm, .mp4, .mkv, and .mov. Anything else won't appear. When you pick a file and run, it resolves the annotated path and outputs it as a STRING.

Two implementation details are worth knowing because they affect how the node behaves:

• The dropdown refreshes based on what's actually in the input folder, so you "upload" by dragging a video into ComfyUI/input/ (or using the upload widget) and it shows up in the list.
• It tracks changes by hashing the filename plus the file's modified time, so ComfyUI will re-execute the downstream graph when you swap the video file. Toss in a new version of the same file and the summary re-runs. That's a nice touch for iterating.

The inputs and output

• video - the only input: a dropdown of the supported video files in your input directory. Since the list is built live, there are no fixed choices in the schema; you'll see your files once the node loads.
• video_path - the output: a STRING with the full path to the selected file.

Wire video_path into VCW_VideoSummary and you're done - that's the intended flow, and the pack's demo workflow does exactly that.

Gotchas

• Only the input directory. If your video lives somewhere else, it won't show up. Copy it into ComfyUI/input/ or use the upload widget.
• It never looks at the file's contents. A .mov that's actually a JPEG renamed to fool you will still appear and still get passed downstream, where the model will politely choke.
• No preview. The video is not loaded or shown; you're choosing by filename.

Installing it

With the rest of the pack: ComfyUI Manager, search ComfyUI-VideoChatWrapper, install. Or:

cd ComfyUI/custom_nodes
git clone https://github.com/lebrosoft/ComfyUI-VideoChatWrapper

Restart ComfyUI. This node itself needs nothing special - but its sibling VCW_ModelLoader will happily download a ~15 GB model on first run, so the real "dependency" of the video-chat half is disk space and an internet connection on the first queue.

Honest framing: this is an obscure pack with effectively zero community footprint, and this node is the least interesting part of it. But as a path-provider for the video-chat pipeline it's perfectly adequate - simple, no state to get wrong, and its file-hash change detection is a small quality-of-life win you don't always get from hobby nodes.