Wan 2.2 Continuation Plan (Nukun)

Extending a Wan 2.2 video is the kind of thing you only do because the base clip wasn't long enough, and the temptation is to just keep clicking "extend." But every extension is another 80 frames of decoded video sitting in RAM while the next segment renders - and unlike image generation, you find out the hard way when it OOMs at frame 700. Nukun Wan 2.2 Continuation Plan is the node that does the math before you commit: tell it how many extensions you want, and it reports the total frame count, the real duration, and a conservative estimate of the decoded-frame RAM the whole run will hold.

How it works

It takes your settings object and an extension_count (1–10), then computes the continuation arithmetic. Reading the source: each base segment is segment_frames long, each extension renders another segment but drops its first frame (the duplicate seam where the new segment's first frame equals the previous one's last), so the total is segment_frames + extension_count * (segment_frames - 1). For the pack's balanced-portrait base that's 81 + extension_count * 80 - ten extensions lands at 881 frames and about 5.1 GiB of decoded-frame RAM by the node's estimate. It also reports motion_duration (playback span of actual movement) and container_duration (the slightly longer container length, since there are total_frames frames at fps). The report output adds warnings when you exceed three extensions (identity drift gets real) or when the RAM estimate crosses 6 GiB.

The inputs and outputs that matter

• settings - from NukunWan22VideoSettings; the whole plan derives from it, so it can never disagree with the actual render.
• extension_count - 1 to 10. Above 3, the report starts scolding you, and it has a point.

Outputs worth wiring: plan (the WAN22_CONTINUATION_PLAN object that ContinuationRecord and ContinuationManifest consume), total_frames, motion_duration, container_duration, estimated_frame_ram_gb, plus extension_count and an empty_log_json (the [] starting log for the record loop) and the human-readable report.

The honest take

This node is a spreadsheet pretending to be a node, and that's a good thing - it's the pack's way of making you confront the cost of a long continuation before you've queued an hour of sampling. The RAM figure is conservative (decoded frames, not the model), so treat it as a floor for what a full run can hold. If you only ever do single-shot videos you don't need it. If you're chasing a two-minute Wan clip, it's the difference between a planned run and a mid-run OOM.

Installing it

Part of Nukun_ComfyUI_Nodes - ComfyUI Manager (search "Nukun") or:

cd ComfyUI/custom_nodes
git clone https://github.com/OnekoSL/Nukun_ComfyUI_Nodes

Restart ComfyUI. Only numpy, Pillow, scipy, and PyWavelets are required - no models, no ffmpeg installs. If the plan errors with "invalid extension count," check that you're passing a plan from ContinuationPlan into the record/manifest nodes, not a hand-typed dict.