A minute to talk
At the moment, I have gone deep into the training process of models and LORA, it seems that I have become a little closer to understanding how to make a neural network understand which character is in front of it and reproduce it quite accurately (without loss of quality and overtraining). Still a work in progress, but the results are already better than before, I hope with time I will understand how to implement this process as accurately as possible, I have read so much in the last few days that I think my brain is about to explode, but I hope it is worth it.
It seems to me that the difficulty of understanding SDXL is directly proportional to the duration of the training (about 15-20 times longer than 1.5)
talk
training
ArtFunkGar
Keep up the learning! And it is doubly cool if you post about what you're picking up. The nuances of how this work is accomplished is pretty interesting to me.
Jun 07 2024 17:11
OneRing (aka GirlsAi)
ArtFunkGar, first I need to learn this, I still haven’t fully figured out what the learning curve depends on, but the results that I learned are already much better than the previous one, when I understand the very essence of “Block dims, Block alphas, Multires, Additional parameters, Optimizer extra arguments " etc. then I will try to explain how to train models on SDXL, but at the moment I think I am not yet experienced enough.
Jun 07 2024 21:46
