Load Video (URL/Path)

One field in, four outputs out: give it a URL or a local path to a video file, and it hands back the decoded frames, the source frame rate, the audio track, and a video-info object - everything a video-to-video or frame-processing workflow typically needs to get started. The point is the same as everywhere else in this pack: you shouldn't have to manually download a clip and stick it in an input folder before your workflow can use it, especially when you're running somewhere you can't just drag a file into place.

Look at the output shape closely and it's clear this node was built to be a drop-in replacement for the video loader in ComfyUI-VideoHelperSuite (VHS) - the widely-used pack that defines the VHS_VIDEOINFO type this node also outputs. Anything downstream that's expecting VHS's standard video-loader outputs (a Video Combine node putting frames back together, a frame-interpolation node reading fps, anything reading video_info for length or resolution) will accept this node's outputs the same way it would VHS's own. The difference is entirely in how the video gets in: VHS wants a file already sitting in your input folder; this node will fetch it from a URL first.

The one input, four outputs

• url_or_path (required) - a direct link to the video file, or a local path if you already have it. That's the entire input.

Outputs:

• frames (IMAGE) - the decoded video as an image batch, one frame per image, ready for anything downstream that operates frame-by-frame.
• fps (FLOAT) - the source frame rate, useful for wiring into a Video Combine node so your output plays back at the same speed as the input.
• audio (AUDIO) - the extracted audio track, if you need to carry sound through to a re-encoded output.
• video_info (VHS_VIDEOINFO) - metadata about the source clip, in the same format VHS's own loader produces.

Installing it

ComfyUI Manager: search ComfyUI-load-lora-from-url. Manual:

cd ComfyUI/custom_nodes
git clone https://github.com/bollerdominik/ComfyUI-load-lora-from-url

Restart ComfyUI. If you're going to wire this into anything that reads VHS_VIDEOINFO, you'll also want ComfyUI-VideoHelperSuite installed, since that's the pack whose nodes actually consume that type on the other end.

Where people get burned

• A URL that isn't a direct file link. The same trap as the LoRA and upscale-model loaders in this pack: a video page URL (a share link, an embed page) isn't the same as a direct link to the video file itself. If the fetch fails or returns something that isn't a video, check that the URL resolves straight to an .mp4/.webm/etc. file.
• No audio track in the source. If the clip you're loading has no audio (a lot of generated video output doesn't), the audio output will be empty - that's expected, not a bug, but it'll break anything downstream that assumes audio is always present.
• Large or long videos on a run-time fetch. Because the video downloads when the workflow runs rather than being pre-staged, a large file adds real time (and, on a metered connection, real cost) to every run that touches this node - worth downloading once locally if you're going to iterate on the same source clip repeatedly, rather than re-fetching it on every queue.