Working on the dataset for Era V1.
Should be considerably broader subject material, and considerably more realistic. Now have a better base flux model for realistic fantasy (Aeromancy), and a better SDXL (Alchemy) model to train on, and I'll probably train for both (SDXL first most likely)
Considerable effort has been made on this dataset already, it's taken months of compositing, selection, multi-passing, upscaling to acheive a realist dataset covering post-apoc, cyberpunk, medieval fantasy, victorian era, the old west with some decent period accuracy in hairstyles, makeup, objects, lighting, complex backgrounds - as well as a modern set for negative weighting.
In addition, obviously, money spent.
The dataset is maybe 60% AI, but only 40% still feel much like it. That's what I'm working on now, closing the gap for the synthetic stuff.
I'm sure I can train the SDXL one well, and flux, well I've never trained it this heavy before, but I'll give it a crack. Tags likely to be complex - I'll create an article on those eventually, covering everything, and some starter pos/neg prompts for each genre.
One day, should I ever have access to real compute, I'll do some full base model trains, as this dataset is starting to get big enough for that.
