Models from the Coupled Model Intercomparison Project phase 6 (CMIP6) typically struggle to reproduce observed Antarctic sea ice trends, a bias that is substantially alleviated when constraining winds. Here we use wind-nudged simulations from two CMIP models to investigate the influence of clouds on sea ice area (SIA). We find that nudging model winds in coupled simulations towards reanalysis, in addition to improving SIA variability, is crucial to reproduce realistic cloud radiative effect (CRE) and cloud cover. We then unveil a negative relationship between biases in CRE anomalies near the sea ice edge and SIA anomalies, which helps explain the remaining discrepancies between simulated and observed SIA: a positive 1 Wm-2 CRE anomaly bias contributes to a negative 0.43 106 km2 SIA anomaly bias. Finally, we find that most CMIP6 models (10 of 12) show positive trends in CRE anomaly biases, which should amplify SIA decline in response to climate warming.