>>74709417Holy shit, thank you.
It seems to have been trained on 138 images?
From the metadata and super early tests it honestly doesn't seem particularly good.
I also think 1.5 was just better at uniforms by default, so its LoRa performed better.
Still, WAY better than without it.