IAMCCS WAN Relay Or Bypass PURE

Prompt relay is IAMCCS's way of letting each chunk of a shotboard sequence carry its own local prompts into a shared generation context - a per-chunk text splice that keeps a long video's narrative coherent. But not every chunk should relay. Some are pure continuations where the global prompt is fine, and forcing relay on them just invites drift. IAMCCS_WanRelayOrBypassPure ("IAMCCS WAN Relay Or Bypass PURE") is the node that decides, per chunk, whether the relay engages or the pipeline just passes through.

It takes the timeline_plan, the current chunk_index, and the standard model/positive/clip/latent signals, and it outputs the model and positive you'd feed the sampler - either relay-adjusted or untouched, depending on what the timeline says for that chunk. The promptrelay_enabled BOOLEAN output tells you which path it took, and report explains the decision.

How the decision works

The node consults the timeline plan's per-chunk relay data. If the chunk's promptrelay_chunks entry is active (it has local prompts), the relay engages and the positive conditioning is rebuilt to splice those prompts in. If not, you get a clean bypass - the model and conditioning pass through unmodified, exactly as if the node weren't there. The relay_options input (type RELAY_OPTIONS) carries the relay configuration so the node knows how to assemble the spliced conditioning.

The lazy behavior matters here: chunk_index is resolved only when it changes, so in a 40-chunk loop the node doesn't do redundant work on every pass.

Why it exists

Without a per-chunk relay switch, you'd need separate graph branches for "relay chunk" and "no-relay chunk" and a router between them. This collapses that into one node that reads the plan. It's also the natural pair for IAMCCS_WanFLFPairFromTimeline, which exposes the per-chunk relay_local_prompts, relay_segment_lengths, and relay_epsilon - those are the inputs the relay path consumes.

Install

Standard IAMCCS-nodes:

cd ComfyUI/custom_nodes
git clone https://github.com/IAMCCS/IAMCCS-nodes.git

or ComfyUI Manager → search "IAMCCS" → restart. No models, no dependencies. ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8.

The practical gotcha

This node only makes sense inside the PURE shotboard pipeline - feed it a plan from IAMCCS_WanCineInfoPure and it behaves; feed it garbage and promptrelay_enabled just reads false and you get a pass-through you didn't need. So check that output once on a chunk you expect to relay; if it's false when it should be true, the timeline's promptrelay_chunks weren't populated upstream, and that's the bug to chase.