Nifty Apply Lora Stack

If Nifty Lora Stack is the recipe, Nifty Apply Lora Stack is the kitchen. This node takes a NIFTY_LORA_STACK wire, a model, and an optional CLIP, and patches them all together - the "apply" half of the pack's two-piece LoRA design. It exists to separate what you want (the stack definition) from when and where you want it (the application point in the graph).

How it works

The mechanics are identical to the Nifty Lora Loader under the hood: it walks the stack in order and runs each entry through ComfyUI's load_lora_for_models, applying the recorded model and clip strengths. Because each stack entry carries its own clip strength separately from its model strength, one Apply node can handle LoRAs that were designed with a different CLIP weighting than the default 1:1.

Two small behaviors worth knowing:

• Entries with both strengths at 0.0 are skipped.
• A LoRA file that can't be found logs a warning and moves on instead of crashing the run.

Inputs and outputs

• lora_stack - the stack wire from a Nifty Lora Stack node (required).
• model - the base model to apply to (required).
• clip - optional CLIP model; leave empty to patch the model only.
• Outputs: MODEL and CLIP, fully patched, ready for your sampler and text encoder.

Why the split is worth it

The payoff is reuse without duplication. The canonical pattern: build one stack, then drop two or more Apply nodes in different parts of the graph. Maybe you're comparing the same character LoRA loadout across two checkpoint models, or you want to apply the stack to the main model and to a refiner model. Edit the stack once, and every Apply node downstream picks up the change - no more keeping three parallel LoRA chains in sync by hand.

It also plays nicely with the stack-chaining feature: a Nifty Lora Stack can append onto an incoming stack, so you can define a shared base stack and let a per-branch stack layer on top of it, with the Apply node (or nodes) doing the final work wherever the model actually gets sampled.

Installing it

It ships with Nifty Nodes, so install the pack once:

cd ComfyUI/custom_nodes
git clone https://github.com/Stibo/comfyui-nifty-nodes

or search "Nifty Nodes" in ComfyUI Manager, then restart. No extra model files - it loads LoRAs from your existing ComfyUI/models/loras folder. Two install-era notes: the pack is built on the new ComfyUI V3 API, so update ComfyUI if the nodes don't show up; and it's a young pack with little community discussion out there, so treat the README and the node source as the documentation.

Gotchas

The most common mistake is wiring a stack into Apply and keeping an old LoraLoader chain in the same path - you'll get the LoRAs applied twice. Pick one mechanism per branch. Also remember the stack is just data: if the Apply node isn't connected, nothing happens and nothing tells you. When a workflow suddenly looks "wrong," check whether the stack changed under you - the definition lives in one place by design, which is a feature until you forget it's the single source of truth.