LoRA Schedule (WAN, ranged)

Some long videos want one LoRA for the whole run - and some want a character LoRA for the first ten generations, a different style LoRA for the next stretch, and a speed LoRA only on every even generation to keep things snappy. IAMCCS_WanLoRASchedule ("LoRA Schedule (WAN, ranged)") is the node that lets you write that timeline down instead of babysitting it.

It's an authoring surface with a genuinely big canvas: 64 slots, each with a lora_name, a strength (-5 to 5), a preset, and a start/end range. You feed it a generation_index (your loop counter) and it resolves which slots are active for that generation, stacking them onto an always-on default_lora and outputting a single combined LORA.

The presets do most of the work

Each slot has nine scheduling presets so you rarely touch raw ranges:

• all generations - always on.
• gen 0 only / gen 1 onwards - for a distinct opening or everything-after.
• even gens / odd gens - alternating patterns.
• every 2nd gen / every 3rd gen - interleaved effects.
• custom range / manual_range - your own start/end. Crucially, end = -1 means "from this generation onward," the open-ended range you need for long loops you don't know the length of ahead of time.

Multiple active slots stack together - ranges can overlap and layer. So you can have a base character LoRA on all generations and a "night scene" LoRA on gens 5–12 that stacks on top of it.

The inputs that matter

• generation_index - lazy, wired to the loop counter. The whole schedule resolves from this.
• model_type - wan2x, flow, or standard. Pick the key-remapping for your base model; flow is the right call for WAN 2.2.
• default_lora - the always-on stack, passed through on every generation. Optional, but that's where your permanent look lives.
• linx - an existing IAMCCS_WAN_LORA_LINX schedule you want to extend. This is how you chain schedules.

Outputs: lora (combined, ready to apply), active_slots (how many fired), report (which slots matched, with names and strengths), and linx (the resolved schedule object - hand it to IAMCCS_WanLoRARuntimeBridge to execute it loop-safely, or pass it on).

Install

Standard IAMCCS-nodes:

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

or ComfyUI Manager → search "IAMCCS" → restart. No model downloads. ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8.

The honest reality

64 slots is a lot of node. You will not use 60 of them, and that's fine - the widget grid is the tax you pay for never having to touch a schedule mid-render. The report output is your friend when a phase comes out wrong: it logs every active slot and the rule that matched, so a LoRA mysteriously missing on generation 12 is a two-second diagnosis. And note the pack's own guidance: keep the always-on LoRAs in a stack node and use this for the phases - that split keeps the schedule readable.