>>59990812I did figure a sort of workaround. I merged the CosineB model into another one of my models in the checkpoint murder at a super high strength. Like my model + cosineB at WS0.78. It doesn't run into the pollack painting issue and can replicate the initial output without issue.
