Apply LoRA to MODEL (LTX-2, staged) (BETA)

Most LoRA apply nodes are boring in the best way: one model in, one model out, one strength slider. IAMCCS_ModelWithLoRA_LTX2_Staged is deliberately not that. It applies two LTX-2 LoRAs to a model in two separate stages and hands you both intermediate results - because a common LTX-2 pattern is to build up a base model, then branch: one path keeps the stage-1 model for a first-pass draft, the other continues on to stage 2 for the final render.

It's from IAMCCS-nodes, the pack by IAMCCS (Carmine Cristallo Scalzi) that grew out of fixing LoRA loading in WANAnimate-style workflows and now carries a whole LTX-2 section. This one sits in the IAMCCS/LoRA category next to the LTX-2 LoRA Stack nodes. The "BETA" tag is the author being honest - these staged nodes are newer and less battle-tested than the plain Apply LoRA to MODEL (LTX-2).

How it works

Where a normal apply node fuses one LoRA into your model, this one chains two of them and exposes the intermediate state. lora_stage1 gets fused first, producing model_stage1 - that's the draft-quality model, wired to your cheap/fast branch. Then lora_stage2 fuses on top of that, producing model_stage2 - the finished model for your full-quality render. Both outputs are live simultaneously, so one graph can draft with stage-1 and render with stage-2 without reloading or re-applying.

The optional model_stage2 input covers the case where your stage-2 base isn't the stage-1 output - say the stage-2 pass should run on a different checkpoint or a reloaded version of the same one. Leave it unconnected and stage 2 just continues from stage 1, which is the standard setup.

Inputs and outputs

• model (MODEL) - your LTX-2 base model.
• lora_stage1 (LORA) - first LoRA to fuse.
• lora_stage2 (LORA) - second LoRA to fuse on top.
• model_stage2 (MODEL, optional) - override base for stage 2; leave empty to chain from stage 1.

Outputs:

• model_stage1 (MODEL) - model with only lora_stage1 applied.
• model_stage2 (MODEL) - model with both LoRAs applied.

Notice there are no strength widgets in the schema - the strengths are baked into the LORA objects you feed it (from an LTX-2 LoRA loader or one of the IAMCCS LoRA Stack nodes). If you want per-stage strength control, stack from a LoRA Stack (LTX-2, staged) upstream instead.

How to install

Part of IAMCCS-nodes. ComfyUI Manager: search "IAMCCS". Or manually:

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

Restart ComfyUI after cloning. The pack wants ComfyUI ≥ 0.3.0, Python ≥ 3.12, Torch ≥ 2.8. No model files ship with this node - LTX-2 checkpoint, VAE, and your LoRAs all come from elsewhere.

Gotchas

The BETA tag is doing real work here: staged nodes in this pack have shifted between releases, so if a saved workflow won't load, updating the pack (Manager's update button, or git pull) is the first fix. And think about what stage-2 chaining means for VRAM - you're holding two patched models' worth of state in the graph; on a tight card, run the branches sequentially and let the graph unload the unused stage. The whole staged pattern is an LTX-2 workflow nicety, not a Wan thing - if you're on Wan 2.2, the pack's runtime bridge or WAN LoRA stack is the better fit.