Nodes/Nynxz/Apply LoRA
ComfyUI Node

Apply LoRA

Apply a shared LoRA stack to any model + CLIP pair

By Nynxz·Created about a month ago·Updated 29 days ago· 5
Apply LoRA
  • model
  • clip
  • lora_stack
  • model
  • clip

Apply LoRA is the consumer end of this pack's headless LoRA setup. You don't build the stack on it - you feed it a stack from a LoRA Picker, plus a MODEL and CLIP, and it returns both patched. It's the "apply this bundle here" node, and its whole reason for existing is that one picker can feed many Apply nodes, so the same stack reaches several places in your graph from a single definition.

The contract

It's a small node and it should stay that way. Three required inputs: model, clip, and lora_stack - that last one has a tooltip that just says "Connect a Nynxz LoRA Picker," which is the entire contract. The output is model and clip, both patched, ready for your CLIP text encode and your KSampler. Because the stack input is the pack's own NYNXZ_EXP_LORA_STACK_DATA type, you literally cannot wire the wrong thing into it - a stock ComfyUI LoRA loader won't mate with it.

The mental model: Picker = the definition of a stack (search, thumbnails, bookmarks, per-row on/off and strength), Apply = the instantiation. Change a strength on the Picker and every Apply node consuming it picks it up on the next run - which is exactly the "one authoritative source" pattern that keeps a multi-LoRA graph from turning into a web of hand-edited loaders.

Where it shines (and where it doesn't)

The win case: the same character stack applied to two checkpoints being compared, or applied at both stages of a two-pass workflow so the LoRAs don't drop out after the first pass. The non-win case: a single-model graph, where the plain LoRA Loader is fewer nodes and does the same job. If you find yourself with one Picker and one Apply, you're adding indirection for no reason - the loader would've been the simpler read.

One gotcha worth naming: since the Apply node carries no stack of its own, the state lives entirely upstream. If you load a shared workflow and the Picker isn't connected, the Apply node has nothing to do and your model comes out unpatched - the graph won't scream at you, it'll just quietly render with no LoRAs. Check the Picker when a stack "stops working."

Installing it

cd ComfyUI/custom_nodes
git clone https://github.com/Nynxz/ComfyUI-NynxzExperimental

or search "Nynxz" in ComfyUI Manager, then restart. No pack-level dependencies; your LoRAs live in your existing models folder. It's experimental, so expect the schema to be a moving target while the pack matures.

CategoryNynxz/LoRA

Inputs (3)

NameTypeDefaultDescription
modelMODEL
clipCLIP
lora_stackNYNXZ_EXP_LORA_STACK_DATAConnect a Nynxz LoRA Picker

Outputs (2)

NameTypeDescription
modelMODEL
clipCLIP