tag:1girl,dark skin, muscular, grey hair,long hair,animal ears, (eyepatch:1.2),red eyes,red choker, large breasts, scar on face,gloves,bare shoulders,cleavage,bandeau,underwear,fur-trimmed jacket,scar on stomach,open belt,green pants,open pants,panties, tail,
有事加群：188546326
Description
FAQ
Comments (5)
I'm curious why you train your LoRAs to require such a long string of descriptive tags instead of a short trigger word or phrase, like "Ghislaine Dedoldia". Having to describe the character every single time, and remember all those tags, is tedious to say the least. For this LoRA, that's 23 tags!
There are two reasons. The first reason is that the more tags there are, the more flexible the character image is, including changing her clothing, her hair color and length, etc. Some users have this demand. The second reason is that when you use a single trigger word to draw a picture, when you use different large models, they will lose different features. For example, if the anime character you train has a tail, it will be difficult to appear in the real model. Sometimes you still need to use the tail tag to reproduce. These are two groups of users, users with single trigger words will be more, but users with complex trigger words will use your model more carefully. For a character, if someone has already made lora with a single trigger word, I'm more inclined to train the second one.
In addition, I use the nai model as the training model. If you make a picture with a large model close to nai, you don’t need to print all the 23 tags, but different large models have different differences. You can observe which feature it forgets , select supplement from my tag
Oh, and I'll add that you can even use my different lora characters at the same time to replace their outfits. For example, https://civitai.com/models/87290/hakurei-reimu-touhou-project
@kozue Okay, thank you. I understand well now. That was a good explanation. I completely understand your reasoning. Since I haven't made any LoRAs, I don't quite know the details of things like that, but my understanding was that for things that don't really change (eye and hair color for example), you'd want those to be included in the single "trigger" and not specify them separately, but for things that do (clothing), you'd want to specify the tags specifically for them. But I understand now why it's good to include the tags for all parts of the character's appearance.