Wan 2.2 Continuation Manifest (Nukun)

The last node in a continuation run has to answer one question: "what exactly did we just make?" Nukun Wan 2.2 Continuation Manifest is that answer. It takes the base run's manifest, the continuation plan, and the accumulated per-iteration log, and merges them into one final manifest plus a filename prefix - the single record you keep when the run is done. Everything before it (plan, record, segment store) built the pieces; this node bolts them into a whole.

How it works

Three inputs in, two outputs out. base_manifest_json is the JSON from NukunWan22RunManifest (your base run's model, settings, seeds, and sampling block). plan is the WAN22_CONTINUATION_PLAN from NukunWan22ContinuationPlan. log_json is the accumulated array from NukunWan22ContinuationRecord - one entry per iteration, holding that segment's captions, prompts, and three seeds. The node combines them into a single manifest_json that documents the base run and every extension, then derives a filename-safe filename_prefix so the saved video and the manifest agree on what they are. The result is the kind of file you can drop in a folder next to the MP4 and reconstruct the whole run from months later.

The inputs that matter

• base_manifest_json - starts as {} if you didn't build a base manifest; wire the real one from NukunWan22RunManifest and the final record actually means something.
• plan - the continuation plan; supplies the extension count and frame totals that frame the log.
• log_json - the per-iteration records from ContinuationRecord.

Outputs: manifest_json (the complete record) and filename_prefix (feed it to your video save/filename node).

The honest take

Like its siblings, this is a bookkeeping node - no rendering, no model loading, no magic. Its value is cumulative: a single-manifest record is what turns a sprawling continuation experiment into something you can audit and reproduce. It's also the natural place to end the pipeline: VideoSettings → TI2VLatent → RunManifest → sampler → SegmentStore for the base, ContinuationPlan → ContinuationRecord → SegmentStore per extension, then this node and FrameSequenceAssembler to close it out. If your workflow is already producing good videos, adding this node costs nothing and gives you the receipt.

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. The classic mistake: leaving base_manifest_json at {} and log_json at [] because those are the defaults - then the "final manifest" is empty. Wire the real outputs through and the node earns its keep.