LORA to help bring more kawaii-ness out of your waifus. In my experience, just like the eyes, getting genuine anime pouts is hard for the AI to get right, or at least in a way that makes it look neat. With this LORA, you should be able to generate pouting characters with relative ease. In addition, you can also add squiggle clouds (denotes annoyance) and anger veins to further emphasize the pout, and even put them in speech bubbles.
However, I would recommend that you mostly only try getting pout faces. Squiggles work alright with the proper settings, which would be with the lora slider at .55 or .5, CFG at 10, and the token/tag trigger emphasized to 1.3. I managed to get consistent results through 3 different models with these settings, until the fourth.
As for anger vein symbols, you're free to try to get them to generate right, but personally, I think that has to be a separate Lora all on its own. Occasionally, it gets it right, but most times the AI will attempt to turn it into a sun symbol or star or something completely unrelated. And if it's not a spoken anger vein, then, bc this symbol is usually drawn on a character's head, it will sometimes get turned into a hair accessory. Anger veins are scuffed in this model, so I suggest simply sticking to squiggles and spoken squiggles, they mean they same thing anyway.
Add this to the negatives, or you'll get several symbols around the character:
(multiple speech bubbles, multiple squiggles, more than 1 squiggle, multiple anger vein, more than 1 anger vein:1.4)
168 Images
Based on NAI, tested across multiple models.
Inpaint testing is inconclusive. My UI was fucking with me and I seriously didn't feel like restarting everything. I imagine getting a pout-shaped mouth to generate shouldn't be too hard, and the symbols I really would discourage you from trying.
Description
FAQ
Comments (4)
After I saw a few of your Lora models on facial expressions, I feel they are really impressive. Can I ask you about the details of your training? In my personal Lora training, I often have the problem that my expressions disappear when I draw a big picture of the character, but not when I draw the face very close to the camera. Have you ever encountered this problem? If you have, how did you solve it?
The AI is notoriously bad at faces the further the distance is at which it tries to generate the character. Usually this is fixed by simply inpainting a better one, where it can focus on that one spot and hires it to a better a look.
@Annawn Great thanks. But I am still try my best to do t2i and get a good face as I want. Anyway, thank you for your advise, i2i will be the option after my recent try fails.
@Annawn after I tried i2i as an alternative solution, what i want finally come out easily. I may continue for that t2i lora training, but your advise really helps. Thank you!
Details
Available On (1 platform)
Same model published on other platforms. May have additional downloads or version variants.



