Hello, my dear Followers!
I've been silent since December but I'm back today with the biggest update as of yet :)
Under the 2025-02-20 folders (and -23 for request) you will find that many models:
LyCORIS: 52 models
Flux: 117 models
Embeddings: 182 models
Small Lora: 393 models
Request: Iced Mocha
I have not been training many new SD 1.5 models lately and I think I won't train with the exceptions of coffee or very special requests.
I've started using flux more and more and when I compare the samples between those two - I kinda like the base results better on Flux. SD 1.5 can generate great things as we all know, but when it comes to capturing real likeness - I feel like multiple trainings of the same person in Flux yield better results than multiple trainings of the same person in SD 1.5
On top of that Flux seem to be the superior model when it comes to prompting (and text) and the support still grows.
I will be training more flux models so you can expect some updates from time to time. The thing is I've met Someone Special and understandably I have less time for my AI hobbies so I can't guarantee any timelines ( as if I was ever punctual in the past :P )
Also just before that I stumbled upon the Hunyuan video models and noticed that you can also train Loras. That is something I want to investigate too, but again, no idea on the timelines :)
Enjoy the models :)
as usual, the trigger is 'sks woman' / 'sks person/man' for all models (except embeddings of course), for Flux this is kinda funny thing because even though it was trained on 'sks', it doesn't really require it.
Cheers!