Everyone's favourite sith lord / dumbass himbo.
Jar Jar is probably my favorite joke character, but I also kinda feel he's a great way to test out LoRA training on different models as his character design is different enough to push the capabilities of the trainer.
Description
This was a failed attempt at using only movie screenshots / 3d models of jar jar to achieve a more accurate lora. However, the quality of the movie screenshots are ASS as the cgi in the movies are very dated. Only use this if you want to mess around a bit.
FAQ
Comments (5)
"The ability to speak does not make you intelligent" - Quigonn Jinn
Hear meesa out
I keep running into an issue with extra eyes on the face.
Hey. Yeah I suspect that's due to illustrious sometimes thinking his eyes are actually ears, it seems to stick around no matter how much I train the lora. Unfortunately I don't have any perfect solutions, but I can suggest these 3 things.
1. Lessening or removing any tags related to emotions or eyes, like "shocked" or "looking at viewer"
2. Trying a different checkpoint or using less style loras if you're using those.
3. Increasing the weight of the lora.
Some more info on the Anima version. Recently went on a trip and decided to see what happens if I left it training non-stop for a few days, this is the result. It's about 3x more trained than what I'd usually do with a LoRA like this. Seems to have had some interesting effects. The anatomy is more consistent than the Illustrious version, but the Bananadaboi style is heavily baked in. So overall I'm not all that happy but it's better than nothing. I've added details in the version description for how to lessen that styling issue.