Earngenix Logo
Skip to main content

Glossary · ComfyUI Basics

What Is a Trigger Word in ComfyUI? (Simple Answer)

By Earngenix Team ·

⚡ Quick Answer

A trigger word is a specific word or short phrase you must type into your prompt to switch on a LoRA, LyCORIS, Dreambooth checkpoint, or embedding's trained effect. Leave it out, and the model has no signal to apply most of what it learned.

Think of it as a password: the file is loaded and ready, but its special trained look only shows up once you say the exact word its creator chose.

Why Trigger Words Exist

When someone trains a LoRA or checkpoint, they usually pair their training images with a rare, made-up word — something like "ohwx" or "sks" — instead of a real word the model already understands. This stops the new training from clashing with or overwriting what the model already knows about common words, and gives you a reliable way to turn the effect on and off.

Where You'll See It in ComfyUI

You type the trigger word directly into the CLIP Text Encodenode's text box, alongside the rest of your normal prompt. It doesn't need to go first or last — just needs to be present somewhere in the text that reaches the model.

Quick Example

A LoRA's Civitai page lists the trigger word as ohwx man. Your prompt would then read: "ohwx man wearing a red jacket, standing in a forest, golden hour lighting" — the trigger word activates the trained look, and the rest of the prompt describes the scene as normal.

Rule of thumb: if a LoRA doesn't seem to be doing anything, check its model page for a trigger word before assuming the file is broken.

Common Mix-Up: Trigger Word vs Regular Prompt Word

A trigger word is specific to one trained file and means nothing to the base checkpoint on its own. A regular prompt word (like "forest" or "red jacket") is just normal descriptive language the model already understands, with or without any LoRA loaded.

Frequently Asked Questions

The LoRA or checkpoint is still applied at whatever strength you set, but its trained effect often shows up weakly or not at all, since the model was never told to activate it. You’ll usually notice the output looking like the base model with little of the trained style.

Check the model’s page on Civitai or Hugging Face — creators almost always list it under the description or "Trigger Words" section. Some LoRA files also store it in their metadata, which certain ComfyUI nodes can read and display for you.

Yes. Some LoRAs use several trigger words for different aspects — one for a character’s face, another for a specific pose or clothing style — and list each one separately on the model page.

No. Many general-purpose checkpoints and some LoRAs are trained to respond to natural, descriptive language instead, with no special word required. Always check the model card to be sure.

See It In Action

Ready to load your first LoRA?

Our full guide covers loading, stacking, and finding trigger words.

Published: 2026-09-09 · Last updated: 2026-09-09

Discussion

Join the discussion

Sign in to leave a comment or reply

💬

No comments yet

Be the first to share your thoughts!