Lora Trigger Loader

Every LoRA that was trained to represent a character or style has a trigger word, and half the time it's buried in the model card - and then you forget to type it and wonder why the LoRA does nothing. Lora Trigger Loader is the node that keeps the trigger attached to the LoRA, so it travels with the file spec instead of living in your memory. On its own it loads nothing; it's pure config, the input half of the pack's LoRA stack.

What it takes

• lora_name - the LoRA file from ComfyUI/models/loras. The dropdown is searchable, which matters once your loras folder has a hundred files.
• strength_model / strength_clip - the two strengths, applied to the diffusion model and the CLIP text encoder respectively. Both can be negative, which is occasionally what you want (a LoRA used to subtract a style).
• trigger (optional) - the trigger word(s) for this LoRA, or empty if it needs none.

Its single output is a lora spec of type KINBURG_LORA - nothing is loaded or applied at this stage. That's deliberate: it's a config node, and the spec gets handed to the accumulator that does the real work.

Why the separation is smart

LoRA stacking in ComfyUI core is a chain of LoraLoader nodes, each one a full apply step - five LoRAs is five loader nodes in a row, each adding another place to get the strengths or the order wrong. The trigger word, meanwhile, lives nowhere in the graph at all; it's a prompt token you have to remember. This pack's answer is two nodes: LoraTriggerLoader carries the spec (file + strengths + trigger) and Lora Unlim Accumulator does the apply. One loader per LoRA, all feeding one accumulator, trigger words appended to the prompt automatically.

The trigger-word problem is real and old: training guidance has hammered for years that triggers should be rare tokens to avoid colliding with real words, and a character LoRA without its trigger is mostly a no-op. If the pack's design only did one thing, this is the one - it makes the trigger a first-class citizen of the graph instead of something you type by hand every time.

Install

Part of the Kinburg-Nodes pack - ComfyUI Manager (search "Kinburg-Nodes"), or cd ComfyUI/custom_nodes && git clone https://github.com/Kinburg/Kinburg-Nodes, restart. No dependencies, no models to download; the only prerequisite is your LoRAs sitting in ComfyUI/models/loras. Standalone it does nothing visible - wire its lora output into Lora Unlim Accumulator and it starts earning its keep. If you're stacking more than two LoRAs or you keep forgetting triggers, this pair is the cleaner way to run your model.