UPDATE: Pony version is now up. Trigger word and tag prompts are all still the same. I'll only have 1 image example for the Pony version so I can save my buzz for other future models.
Felicia from the Shrek 5 teaser trailer. This is out of my norm with the models I make, but I wanted to make this particular one since I like this character design and it was a pretty quick one to do too. Of course, the image dataset is low with only 9 screenshots I took from the small trailer, so keep that in mind. I am also aware that in the trailer she has brown eyes, but she's supposed to have blue eyes. Maybe that was a mess-up on Dreamwork's end lol.
Positive Prompts: FeliciaS5, female humanoid, ogre, ogre ears, brown hair tie, brown eyebrows, amber hair, braid, long hair, brown eyes, black pupils, freckles on face, silver nose piercing, green lipstick, green body, green skin, slightly chubby female
For the outfit: necklace, green crop top, belt, belt pouch, brown bottomwear
Put these in the following for more accuracy on the outfit and look in the negative prompts:
cheek piercing, human ears, humanoid ears, earlobes, rainbow mouth
For some reason there's some really bad artifacting issues with this model especially around her mouth if you have it closed without her teeth showing. If that happens, then use the "hi-res fix" function on the image you generated and it will almost always fix the issue. You can also combine it with other art style models and it'll help reduce that, but using the model solo will make it more likely for that artifacting to happen. 0.8 on the weight has given me best results since if you increase it higher than that, then the artifacting of the images you generate will more than likely show up stronger. I assume it's because of the trailer's quality I trained the images on. Even though it's at 1080p it still has some bad blurry pixels in it. If there's ever a new trailer with a much better video render quality to train it, then I'll go update it. Also, the necklace doesn't come out correct most of the time, but it can appear mostly accurate on rare occasions. Finally, I know she doesn't wear a crop top and it's more of a tucked in shirt, but for whatever reason, the training recognizes her topwear best if I put "crop top" in it and not "shirt." Again, probably due to the low image count I can only get from the trailer, but it's better than nothing for now.