Works decently
er_sde/beta and res_multistep/sgm_uniform used in samples
Description
should be overall better but use whichever you want, adding smooth shading to prompt helps but i dont know if the supposed fact that there's lora incompatibility with base trained and sft checkpoints, i just cant get it fully to look like even with ridiculous step amounts on training and to be honest im too fatigued to try and make it any better