We propose a new framework for estimating the daily, time-varying, covariances among stocks that incorporates the intraday high-frequency information through realized volatility measures. We work with dynamic stochastic volatility factor models and Cholesky stochastic volatility models with priors that shrink posterior estimates in the direction of the observed realized covariance kernels. The use of this additional information helps stabilizing the variation of high-dimensional covariance estimates and allows the construction of more realistic and hence complex models. We explore the importance of using high-frequency data by comparing the behavior of portfolio strategies based on the proposed estimates.