Apply Scheduled WAN LoRA From Conditioning

When you don't want to wire the loop index into your model branch

The IAMCCS_WanLoRAHookSchedule sibling needs a generation_index wired in - usually from forLoopStart.index - to know which LoRAs to apply on each pass. That's clean, but it has a catch: in some workflows, routing the loop index into the MODEL branch (the one that patches weights) is awkward or creates validation cycles. This node removes the need. It resolves the effective generation index from the conditioning instead of from a loop counter, then applies the scheduled LoRAs to the model.

The mechanism, straight from the source: it uses the same 64-slot schedule machinery as IAMCCS_WanLoRASchedule, but instead of a generation_index input, the effective index comes from metadata emitted by the pack's IAMCCS_WanIndexedPromptEncode. Because that prompt encoder tags each conditioning with its segment/generation index, this node's index is always synchronized with the prompt that actually reaches the sampler - even when the graph wiring doesn't make the loop index easy to route.

The interface

Required: model, positive, negative (so it can pass both conditioning and the patched model through in one place), plus the full slot schedule - log_prefix, model_type, and per-slot lora_name, strength, preset, start, end for up to 64 slots. Same presets as the hook node: all generations, gen 0 only, gen 1 onwards, even/odd, every-Nth, and custom range/manual_range with end = -1 meaning "onward."

Outputs: the patched model, the passed-through positive/negative, the resolved generation_index (useful for your own bookkeeping), a report of what fired, and a linx contract.

When it's the right tool

Reach for it when your loop's MODEL branch is hard to wire to the index - for example when the sampler lives in a different part of the graph than the loop counter, or when you want one less wire crossing the canvas. The tradeoff is the dependency: it only works when your positive/negative conditioning comes from IAMCCS_WanIndexedPromptEncode (or anything else that stamps the generation metadata the node reads). If your conditioning doesn't carry that metadata, the index will be wrong or default - and a silently wrong index is worse than a visible one, so check the generation_index output on your first run.

Like the rest of the pack it installs via ComfyUI Manager (search "IAMCCS") or git clone https://github.com/IAMCCS/IAMCCS-nodes.git into custom_nodes, then restart. No extra dependencies beyond a working WAN/Flow setup. If you're building simple one-off graphs, the plain hook schedule is easier to understand at a glance; this one earns its keep in big indexed-multigen workflows where prompt and model must stay in lockstep.