LoRA Stack (LTX-2, staged: stage1+stage2) (BETA)

LTX-2 workflows commonly run in two sampling stages - and the community has learned that LoRA strengths that work for the first stage don't necessarily work for the second. The distilled-LoRA testing that people actually did for LTX-2.3 settled on different strengths per stage (0.4 on the first sampler, 0.5 on the second in one well-known test). This node is built for exactly that pattern: it's the 3-slot LTX-2 LoRA stack, but each slot carries two strength values, one per stage, and it emits two LORA bundles out the back.

It's marked BETA in the pack, which is honest - but it's also the node that makes stage-aware LoRA application trivial instead of a mess of duplicated loaders.

How it works

For each of the three LoRA slots you set two strengths:

• strength1_stage1 / strength1_stage2 (and 2_ / 3_ for the other slots).

A slot with "no" or both strengths at 0 is skipped. Then:

• Stage 1 bundle = each slot at its *_stage1 strength.
• Stage 2 bundle = each slot at its *_stage2 strength.

The optional lora input (an existing LORA bundle) is included identically in both stages - handy when you have an always-on base stack plus stage-specific extras. Outputs are two LORA outputs, lora_stage1 and lora_stage2, each fed to its own apply node (IAMCCS_ModelWithLoRA_LTX2_Staged is the natural partner, which outputs two models) or to your two samplers' respective model inputs.

Same conservative key handling as the rest of the family: legacy lora_down/lora_up → lora_A/lora_B, no aggressive remapping.

Inputs and outputs

• lora1/lora2/lora3 - file dropdowns.
• Six strength fields (stage1 + stage2 per slot).
• Optional lora input.

Outputs: lora_stage1 and lora_stage2 (both LORA).

Install

ComfyUI Manager → "IAMCCS", or:

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

Restart. No extra dependencies.

Gotchas

The whole point is that stage strengths differ - if you set them identically you're just using a fancy version of the plain stack. Think of stage 1 as the shaping pass (often stronger) and stage 2 as the refinement/audio-sync pass (often gentler, especially for distilled models where over-strong LoRAs reintroduce artifacts). BETA means the interface may shift between pack versions, so a downloaded workflow built around it might need a slot re-pick after an update. And remember the external lora input goes into both stages - if that's not what you want, split your always-on LoRAs into a separate always-on stack instead.