OneTrainer - My tips for preparing a data set

best canvas sizes are canvas
1024.00 x 1688.00 and above to get best likeness as reduces compression artifices etc.
my onetrainer setting,
www.mediafire....
Ai Program kit,
www.mediafire....
my cuda tool kit installed appears to be v12.2 available here
developer.nvid...
my python version
python 3.10.6
Visual basics - install (note: very very large will take awhile to install and needs a lot of space)
python development
Microsoft development with C++
visual studio extension development
www.mediafire....

[@eccentricworx]

You don't need to flip images in your set manually, in the concepts you'll find an option for onetrainer to flip them randomly, which is very useful to save steps and keep a variety in the data input. When activated, my loras get more symmetric and, for some reason, get cleaner details. I wonder how this adaptive step adjusting thingy that you mention is working, never happened to me, I will look into it, could help to prevent overtraining....

[@AgustinCaniglia1992]

Does this apply for flux as well? About the 1024 resolutions?

[@eccentricworx]

I guess so. I also read somewhere that you still get excellent results in flux with 512px...

[@henrywang4010]

Is the cat real?

[@streamtabulous]

green screen, cat, flames, fan,