santa hat
deerdeer nosedeer glow
Sign In

Can we make it so that all LoRAs have no triggers?

Initially, it seems fine to use trigger words... buuuuut... now that I have been using 16 models and 121 LoRAs, it is becoming way too chaotic, and it is only going to grow rapidly. Especially all the LoRAs that require one trigger, what's the point?

2 Answers

Technically, there is precious little control over how people choose to train, so no.

There is some value in having a keyword, in that it can be positioned and duplicated on top of merely weighted, which might offer finer control.

What you need is extensions which know and add keywords automatically: has a list (though bafflingly it still uses the flawed autoV1 hash which is prone to collisions), uses the aforementioned extension or user's manual keywords to auto add either, and possibly there's the civitai (helper) extension, I haven't kept up with its capabilities.

The addition of loras & variants has been patchwork from the beginning and I'm still hoping to see them simplified like embeddings, but until then you do what you can.

Trigger word is also "emphasis" word for training process. When you're training a LoRA w/o keep tag setting, you're training mostly whole images. Of course, it will train the same object on its own, for example, if all your images has white baseball hat, but it will be less efficient and more style-baking that just using single trigger word.

If you don't want to to lose all trigger words for your LoRAs, you can just keep all notes about LoRAs in new Automatic tab (1.6+), you can open it by hovering over LoRA in LoRA tab.

It has its own interface with different boxes.

And you can use "Notes" to keep all important things here. Or you can add any text with LoRA name, by putting in into "activation text" box.

If we'll talk about training style LoRAs, there is no requirement for trigger word, it's even better not to use it.

Your answer