BerniniR · Load Video (webp/gif/mp4/mov/...)

BerniniR · Load Video turns a clip on disk into the frame batch that BerniniR · Encode Source/Reference expects for video editing (v2v / rv2v). It reads webp and gif with Pillow - zero extra dependencies - and mp4/mov/avi/mkv/webm through whatever video backend you happen to have installed. It's the pack's own little frame loader, and for a beginner it beats wiring up VideoHelperSuite just to get frames in.

The one decision that actually matters lives in its defaults, and it's a trap worth understanding before you hit it.

The inputs

• video - a dropdown listing videos in ComfyUI/input. The upload button handles webp/gif; for mp4 and friends you drop the file into ComfyUI/input yourself and it appears in the list. The README's sample value is source.webp, and that's a fine place to start.
• frame_load_cap (default 33) - maximum frames to load. This is the RAM trap. 0 means "all frames at original resolution," and a long 4K clip is comfortably over 10 GB of RAM. The default 33 keeps memory sane. For a full-length 81-frame edit, set it to 81 (or to the clip's frame count) and pair it with max_side.
• max_side (default 1024) - downscales each frame to this longest side on load. It's the second half of the 4K protection: cap limits count, this limits size per frame.

Remember Bernini editing is heavy by design - the source video is part of the model input, so a 81-frame edit processes roughly double what a plain Wan generation would. Loading 200 frames at 4K "because the clip is nice" is how you OOM before the sampler even starts.

Output

frames (IMAGE) - a [T, H, W, C] batch in 0..1, straight into Encode Source/Reference's source_video input (or anywhere else that eats an image batch).

Install

Shared pack install: ComfyUI Manager, or clone into custom_nodes + pip install -r requirements.txt. The webp/gif path needs nothing extra. The mp4 path needs at least one video backend - see below.

Troubleshooting

• "BerniniRLoadVideo: no pude decodificar ese vídeo" (couldn't decode) - that's the node telling you no video backend is installed. Any of these fixes it, in the pack's own preference order:

  pip install av            # recommended
  pip install opencv-python
  pip install "imageio[ffmpeg]"

• OOM / frozen while loading? frame_load_cap=0 on a long clip will eat all your RAM before the first frame renders. Set a finite cap and lower max_side.
• WebP/GIF animations coming out as one frame? Shouldn't happen with this node - it iterates ImageSequence properly. If you're feeding a static-looking webp that's actually a single frame, it is, in fact, a single frame.
• Already using VideoHelperSuite? Fine to keep it: the README explicitly says VHS_LoadVideo wires into the same source_video input. This node exists so you don't need VHS for the simple cases.