Although crucial for the Earth's climate, clouds are poorly represented in current climate models, which operate at too coarse grid resolutions and rely on convection parameterizations. Thanks to advances in high-performance computing, it is becoming feasible to perform high-resolution climate simulations with explicitly resolved deep convection. The added value of such convection-resolving simulations for the representation of precipitation has already been demonstrated in a number of studies, but assessments about clouds are still rare. In the present study, we analyze the representation of clouds in decade-long convection-resolving climate simulations (2.2-km horizontal grid spacing) over a computational domain with 1,536 x 1,536 x 60 grid points covering Europe and compare it against coarser-resolution convection-parameterizing simulations (12-km horizontal spacing). The simulations have been performed with a version of the COSMO model that runs entirely on graphics processing units. The European Centre for Medium-Range Weather Forecasts Re-Analysis-Interim reanalysis-driven present climate simulations (1999-2008) show that biases in mean summertime cloudiness and top-of-the-atmosphere radiation budget are reduced when convection is resolved instead of parameterized. Especially, the typically underestimated midtropospheric cloud layer is enhanced, thanks to stronger vertical exchange. Future climate simulations (2079-2088) conducted using pseudo global warming experiments for a Representative Concentration Pathway 8.5 scenario show a predominating reduction in low-level and midlevel cloud cover fraction and an increase in cloud top height, implying positive cloud-amount and cloud-height feedbacks. These positive feedbacks are only partly compensated by the negative cloud-thickness feedback. Although the simulations exhibit substantial differences in terms of clouds in the present climate, the simulated cloud feedbacks are similar between the 2.2- and 12-km models.