>>73585357Yeah, sorry about the size. When I try for new base models (in this case with new SDXL-friendly parameters), I crank up the dimensionality to eliminate a variable from my testing. It takes a while to train epochs.
I've been getting a lot of request to do Pony (and I was reluctant at first, thinking Pony might be a flavour of the month fad), but my experiment worked out, so I published it for their benefit. I'm hoping to train a smaller version in the future. And then maybe add some more data. It takes time to tag at the accuracy I need.