FunPack Continue Video

FunPack Continue Video is one of those nodes you'll use constantly and never think about twice. It takes a frame batch, cuts off the last frame_count frames, and outputs just that tail. That's the entire job. In a pack that otherwise wants to rewrite your workflow around an AI director, this one is a pure utility - and it's exactly the piece you need when you're chaining video generations.

What it's for

Every "make a longer video" workflow hits the same wall: video models generate a fixed number of frames, and the model has no memory of what came before. The standard trick is to feed the ending of clip A into the start of clip B so the motion continues. But you don't want the whole clip - you want the last few frames. That's this node's one trick: CONTINUED comes out as the trailing frame_count frames, ready to feed an image-to-video or continuation step.

It's the sibling of FunPack StoryMem Last Frame Extractor (which also hands you a final frame for continuity), but with a different purpose. That node splits output into a single anchor frame plus a motion tail. This one just slices whatever tail you ask for. If your next node only needs one frame, set frame_count to 1 and you've got an anchor; if it wants motion context, ask for 5–8.

Inputs and output

• images (IMAGE) - your clip.
• frame_count (INT, 1–9999) - how many trailing frames to keep.
• CONTINUED (IMAGE) - the last frame_count frames.

There's one footgun baked into the source: if frame_count is larger than the total frame count, the node raises an error rather than silently clamping. So if you're feeding in variable-length clips, either keep frame_count conservative or clamp it upstream. It's a deliberate fail-fast design - better a clear error than a hallucinated batch.

Install

It ships in ComfyUI-FunPack, so installation is the pack install:

cd ComfyUI/custom_nodes
git clone https://github.com/olivv-cs/ComfyUI-FunPack
pip install -r requirements.txt

or search "ComfyUI-FunPack" in ComfyUI Manager. No models, no optional dependencies, no GPU involvement - it's a tensor slice, pure and simple.

Where it sits in a real workflow

Typical usage: your KSampler outputs clip A → decode → this node takes the last few frames → those go into the i2v input of the next generation. Combine it with the Scene Chain Sampler or FunPack Video Stitch and you've got the raw material for a longer montage. It's also handy inside the StoryMem loop, where a short tail of recent motion frames gets reused as context for the next shot. It won't fix a model that doesn't know how to use reference frames - but for the plumbing that moves a clip's ending into the next clip's beginning, it does exactly what it says.