Video Chunk ⚡🅡🅞🅣🅘

Long video workflows in ComfyUI hit a VRAM ceiling before they hit anything else - running a sampler or a VAE encode/decode pass over a couple hundred frames at once is a fast way to OOM even on a beefy card, and it's a common enough complaint in the community that it's basically a rite of passage for anyone moving from single-image work into video. VideoChunk is a small, purpose-built utility for the obvious fix: don't process the whole batch at once.

How it works

It takes an image batch - your video frames - and a chunk_size (default 4, minimum 1), and groups the batch into chunks of that size for downstream processing rather than handing everything through in one pass. It's a plumbing node, not a documented subsystem - the pack's README doesn't go deep on this one specifically, so beyond what the schema tells you, the honest read is: treat it as the tool for keeping your batch size within whatever your GPU can comfortably run a sampler or VAE pass on, and check the node's own ? docs icon in the corner for any edge-case behavior beyond what's covered here.

The inputs and outputs that matter

Just two things: image, the frame batch you're feeding in, and chunk_size, an integer with a floor of 1 and a default of 4. The output is IMAGE - your frames, chunked.

How to install it

Via ComfyUI Manager, search "RyanOnTheInside." Manually:

cd ComfyUI/custom_nodes
git clone https://github.com/ryanontheinside/ComfyUI_RyanOnTheInside.git
cd ComfyUI_RyanOnTheInside
pip install -r requirements.txt

Restart ComfyUI. No model downloads needed for this node - it's plain tensor slicing, no dependency beyond what the rest of the pack already pulls in.

Common issues & troubleshooting

The failure mode you'll actually hit isn't from this node malfunctioning - it's forgetting to use it in the first place. If you're feeding a long clip straight into a heavy sampler or VAE node and hitting a CUDA out-of-memory error, that's your cue to insert VideoChunk (or lower chunk_size if you're already using it) rather than fighting the sampler's own settings.

Worth knowing before you rely on it for everything: chunking breaks continuity across the boundary between chunks. Anything in your graph that accumulates state frame-to-frame - a particle simulation, a spline modulator tracking motion over time - can produce a visible seam where one chunk ends and the next begins, because those nodes were never shown the frames on the other side of the cut. That's a real cost of chunking video, not a bug specific to this node, and it's the trade-off you're making for staying inside your VRAM budget.