(Nobody has made a Lora for the Smilfy version of Snuffy)
Fine... I'll do it myself...
***
This is my first trained Lora, trained on 92 images.
"snuffy" is the activation prompt, but putting things such as "belt, bikini, cow print, asymmetrical legwear" will get gens to look more accurate clothing wise.
Example images were generated using Mokoumix. Lora name in example image prompt might be different to the download file, that's because civitai renames it.
Description
OG version
FAQ
Comments (4)
I'd like to know, how did you train the LoRa ?
By that I'm not just talking about the settings, but mostly the image captioning. I understood how to do it for style lora but not characters.
Is it just describing the entirety of the image except the character and their defining traits ?
yeah thats pretty much it, the trainer identifies whats there and defines it as not part of the model since you dont tag the character traits, and you would put the trigger word, the character name, as the first tag on every caption
You really only need 2 - 3 defining words. Character name, Outfit, and traits. The simpler the better. For multiple outfits, it's best to not add the name of the character.
from what i discovered from training (and what i read online), when tagging images, what you more or less don't tag gets incorporated into the Lora. i.e, by not tagging her hair color, eye color, ears and tail, makes it get incorporated into the trigger tag. So the trigger tag "Snuffy" becomes all of those traits