santa hat
deerdeer nosedeer glow
Sign In

Ask how to make a good prompt or trigger word for LoRA | Cloth,Style or Background

What I want to ask is how to fix items that have 2 different data sets into one and wear clothes that are the same color, but different models, how do you fix the prompt so that when it's triggered it doesn't appear jumbled or random but is more focused ?

2 Answers

Hi, some people doing something like that. Fox example, we have two different dresses on your character.

You should clean all tags from dataset* related to dress.

And you're just typing something like: "srtdrs" for short dress. and "lngdrs" for long dress.
You can depsribe these dresses separately. With just one tag. Fox example: "dress" for something with short skirt or "gown" with long. You don't need to describe sleeves, cutouts etc if you the same dress on every image in the subset*.

The same for Characters. If you want to do character in original clothes - charnamedef, if you want to dress them in something else - charnamernd. Def for default clothes subset* and Rnd for other random clohes subset* in dataset*.

For style you usually don't want to specify any style-related tags. Styles usually tranined w/o any trigger words.

I'm doing the same for my LoRAs. And it works as it should.

dataset* - all your training images.
subest* - single folder in your traning images.

Using a LoRA that generates a specific outfit will override the individual outfits in a model. So I would find a LoRA with the outfit you want, or train a LoRA with that outfit yourself.

Your answer