The model was trained on the basis of Pony XL.
Trigger word: megumin
Strength: 0.7-0.8
All my generated images were taken through this service.
I'm still a beginner in neural network training, so I could make a lot of mistakes in training the model, in particular, image selection. Since otherwise I used general tips and instructions, and also tried to follow the default settings in training on this service.
As you may have noticed, I try to match the original style of drawing as much as possible, even if it is very clumsy in the original - it feels closer than ultra-realistic art, far from the canon.
Description
FAQ
Comments (3)
Hey dude, great model! most loras ive made range between 500 to 1600 images. if you want some tips or help culling bad images we can talk in dms
ty for sharing your megumin lora, just wanted to know the difference between the 2 loras, an also why clip skip 1 for pony?
The difference is mainly in the number of images she trained on. So megumin045_ponyXL trained on 45, and megumin364_ponyXL, respectively, on 364 images. My decision to use clip 1 was due to the fact that, it seems to me, it is much closer to making the model more consistent with a more "simple" drawing, not loaded with unnecessary details. Naturally, I can make mistakes, so I'm ready for criticism.
For the most part, I now experiment in my free time, which is why I do such, at first glance, not quite logical things.