>>41811774>>41811850 (me)The LoRA works fine when the appropriate tags are pasted in from a dataset, which makes me question if the token is even working as intended (since it won't work unless the other tags are also there), I just don't feel like i've done it right when I have to refer to the dataset to prompt anything from it. I'm using the colab to train if that matters, though I've also enabled shuffling of captions which I assume should have prevented a situation like this.
>>41811986I think the way I was doing it was making a tag with the character's name and adding it to the beginning of the tags, in all of the images with her wearing her default outfit. and making a separate tag on all the images of her wearing the alternate outfit instead of the character name. Is that the wrong way to go about it? I figured prompting the first one would give them their default outfit at least, but the AI gets it mixed up regardless.
