Quoted By:
5 hours to copy and caption 35k images, another 5 hours to cache the latents and all just for a test run to see if it can learn some holos with 1000+ images in a single epoch.
I was initially planning on doing it on 500k images but I think my hopes are crumbling before even starting. Maybe if I did this back in january it would have been worth it.
Anyway, cya in probably another 12 hours to see if it knows who suisei is in a single epoch of lokr training.
