VRGDG_LoadVideos

VRGDG_LoadVideos loads video files from a folder and hands them to ComfyUI as image frame batches. It's the input side of any video-to-video work this pack does - grab your clips, decode them to frames, and feed the frames into a VAE or an enhancement node.

Mechanically it's straightforward: you give it a folder path, it reads the video files inside, and it outputs an IMAGE tensor (the frames, batched along the time axis, the way ComfyUI represents video). What makes it slightly unusual is the trigger input: a wildcard socket that exists so the loader re-executes when something else in the graph fires - which is the pattern this pack uses to re-load per scene or per run without restarting the workflow.

Inputs

• video_folder - path to the directory containing your videos. Default ./videos.
• scene_count - how many videos to load (1–5, default 3). This matches the "one folder of scene clips" structure the music-video workflow writes out to.
• trigger - the wildcard re-run signal.

Output is a single video (IMAGE) socket. That frames tensor goes wherever video frames go in your graph: into a VAEEncode before sampling, into an upscale or enhancement pass, or straight to a SaveVideo-style output node.

Where you'd use it

Two main cases. First, real video-to-video: you shot or generated clips, you want to re-render or enhance them, and this is the cleanest way to pull them in as frames rather than dragging files through a video loader with a dozen decoder settings. Second, and more in the spirit of this pack: the workflow writes out per-scene clips to a folder, and LoadVideos pulls them back in for a second-pass grade or LoRA-strength pass. It's the "read our own output back in" node.

The scene_count cap of 5 is a tell that this is tuned for the author's chunked workflow rather than arbitrary batch video loading - if you need to slurp a hundred clips, look elsewhere; if you need to pull in a handful of scene files per run, this is exactly right.

Install

cd ComfyUI/custom_nodes
git clone https://github.com/vrgamegirl19/comfyui-vrgamedevgirl

Restart ComfyUI; it's under the Video category. ComfyUI Manager: search vrgamedev. Video decoding here leans on imageio, which is one of the three packages in the pack's requirements.txt - run pip install -r requirements.txt from the repo folder if the node errors on import or fails to read a format.

Honest note

Don't assume it handles every container and codec. imageio covers the common formats fine, but exotic codecs can fail silently or refuse to decode - if a clip comes back empty, transcode it to a plain mp4/H.264 first and retry. And the README for this pack is famously sparse about the actual node set, so if trigger behavior feels opaque, that's expected; the wildcard exists so the loader re-fires when upstream changes.