LoRA Trigger Injector

The most annoying part of a LoRA workflow isn't the LoRA - it's the trigger word. The card says "use risu in your prompt," and you either remember to type it or you don't, and when you're stacking six LoRAs you definitely don't. The LoRA Trigger Injector (NoelLoRATriggerInjector) is a small text utility that reads which LoRAs are actually active in your workflow and splices their trigger words into your positive prompt for you. Configure each LoRA's trigger once; from then on, flipping LoRAs on and off in your stack updates the prompt automatically. No more cross-checking CivitAI cards mid-batch.

Keeping track of trigger words across a growing LoRA folder is a real, constant community pain - it's half the reason rgthree's Power Lora Loader reads them off files for you. This node is the "set it and forget it" version of that idea, aimed squarely at the Efficiency Nodes crowd.

How it works

The node sits between your LoRA stacker and the sampler. It takes a LORA_STACK - the exact type emitted by LoRA_Stacker_ED, which is why the pack's README tips its hat to NyaamZ's efficiency-nodes-ED fork - plus a positive_prompt. You define slots: pick a LoRA in lora_1_name and type its trigger words into lora_1_triggers (comma or space separated). On every run it walks the incoming stack, finds entries whose name matches one of your configured slots, collects their triggers, and appends (or prepends) them to the prompt. Then it passes both the stack and the enriched prompt downstream.

Note what it does not do: it never applies a LoRA. It edits text only, and passes your LORA_STACK through untouched, so you still wire the stack from your stacker into whatever actually loads it.

Inputs that matter

• lora_count (default 5, 1–50) - how many slots to show. The frontend (a bundled lti.js) hides and shows slots live, so crank it to 8 and a slot appears.
• dedupe (default on) - skips a trigger that's already in your prompt.
• mode - append (end of prompt) or prepend (start).
• lora_stack + positive_prompt - technically optional, practically the whole point. No stack connected, no injection; you just get a passthrough.
• lora_X_name / lora_X_triggers pairs - your trigger library, per slot.

Outputs: lora_stack (pass-through), positive_prompt (the enriched one), and injected_triggers - a single space-joined string of what got added, handy for a text-display node if you want to see the node working.

Installing it

It's part of the ComfyUI-NoelTextUtil pack, so install that once:

cd ComfyUI/custom_nodes
git clone https://github.com/noelkim12/ComfyUI-NoelTextUtil

then restart ComfyUI - or just search ComfyUI-ComfyUI-NoelTextUtil in ComfyUI Manager (the title is double-prefixed, don't let that throw you) and click Install. Zero model downloads and no Python dependencies beyond ComfyUI itself; the only thing to install separately is a stacker that emits LORA_STACK, i.e. Efficiency Nodes (ED) or a compatible fork.

Where people get burned

• Exact-filename matching. The dropdown is populated from your loras folder, and matching compares the stack entry's name against it exactly. A renamed file or a stack built from a different name string silently produces no trigger - and with dedupe on you won't even see a duplicate to tip you off.
• The trailing comma. The source appends a comma to the output prompt unconditionally, and injected_triggers is a plain string, not a list. Both are harmless in practice; just don't expect a typed list out of the third output.
• It's niche by design. If you're on rgthree's Power Lora Loader, which already pulls trigger words off files and CivitAI, this node is redundant. Reach for it when you're living inside LoRA stackers and want prompts to stay honest with the stack.