An experimental LoRA trained on images of mesugaki via the NAI model. Does this actually do anything? I can't be certain, but personal testing has shown this LoRA to be a mild quality enhancer, give characters sharper eyes, and make them lewder and smugger (or maybe I'm just biased as the model's creator).
Suggested resources courtesy of mangaka92 (check the comments below for more details).
Prefer textual inversion?
https://civarchive.com/models/108761/mesugaki-textual-inversion
You can also find me on pixai.
Description
FAQ
Comments (5)
It's a decent start, and a concept that definitely needs a high quality LoRA, so thanks for your contribution.
See my comment. It works really well if you combine it with a few other LoRAs.
It seems to do a really good job of making the character's appearance look like a "mesugaki" character, such as capturing their proportions (e.g. height, size of eyes, etc.), but seems like it struggles to capture the "smug" look, sometimes. Using "(smirk, smug)" as triggers seems to work just as well with or without the LoRA.
However, there's three other LoRAs that, when combined with this one, work really well together.
For "smug" expression:
https://civitai.com/models/95220
For "ojou-sama" pose (add trigger "hand_over_own_mouth"):
https://civitai.com/models/51582
For "half-closed eyes":
https://civitai.com/models/35808
(Optional) For "shortstack" body type:
https://civitai.com/models/66679
I'd highly recommend adding each of these to the suggested resources. I've added some samples of the effects of adding this LoRA to the other three, with each LoRA used in the image description (one of them is currently in moderator-review hell, so maybe it will appear a few months from now). This mesugaki LoRA does a good job tying all of the pieces together.
Thank you for your in depth testing and review! I'll definitely play around with this later.
@AInonymous Not a problem! Thanks for your work!