Saw some redditor experiments with flow-matching sd 1.5.
Is flow-matching to blame for lack of variability between seeds?