FL Wan Video Blender

WAN's native generation window is 81 frames, and it hasn't moved since the model shipped. If you want anything longer than that, you're stitching multiple generations together, and FL_WanVideoBlender is Fill-Nodes' simple answer to the stitching problem: feed it two WAN-generated clips and it cross-fades between them over a configurable number of overlap frames, auto-resizing the second clip to match the first so you don't have to pre-match dimensions yourself.

It's worth knowing this sits inside a bigger pattern the whole WAN ecosystem has been building around that fixed 81-frame ceiling - chunk-and-restitch workflows, error-recycling extension LoRAs, single-node chunkers that hide the seam-management work from you. FL_WanVideoBlender is the manual, no-frills end of that spectrum: a straight cross-fade, not a learned blending model. That's a feature, not a shortcoming - it's fast, predictable, and doesn't need anything beyond the two clips you already generated.

The inputs are minimal. video_1 and video_2 are your two clips (as image batches, WAN's usual output format). overlap_frames (1–1000, default 10) sets how many frames at the tail of video_1 and the head of video_2 get blended together into the transition - everything outside that window stays untouched. The single output, blended_video_frames, is the combined result: video_1 up to the overlap point, the cross-faded transition, then video_2 continuing on.

Worth flagging up front: this node lives in the pack's own WIP category (🏵️Fill Nodes/WIP), so treat it as functional-but-not-fully-polished rather than a production-hardened pipeline stage. If a straight two-clip cross-fade isn't quite what you need, the same pack ships FL_WanVideoBlender3Way for Input → Continuation → End three-clip chains, and a separate variant purpose-built for clips produced by FL_WanVideoContinue that strips the embedded overlap frames automatically so they don't get duplicated.

Installing it is the standard Fill-Nodes pattern - search "Fill-Nodes" in ComfyUI Manager, or clone it directly:

cd ComfyUI/custom_nodes
git clone https://github.com/filliptm/ComfyUI_Fill-Nodes

• then restart ComfyUI. No extra dependencies beyond the pack's usual image-tensor handling.

Where people get burned. Since video_2 gets auto-resized to match video_1 rather than requiring you to pre-match resolutions, mismatched aspect ratios won't error out - they'll just get stretched to fit, which can look subtly wrong if your two source clips weren't generated at compatible dimensions to begin with. Match your generation settings between clips up front if pixel geometry matters to you, rather than relying on this node's resize to save you. And overlap_frames needs to actually be smaller than both clips' frame counts for there to be anything to blend - there's a lot of headroom in the parameter (up to 1000), but real WAN chunk-and-restitch workflows typically reuse a modest handful to a couple dozen frames as the overlap window, not hundreds, so keep it proportionate to your actual clip lengths rather than maxing it out.