Cloud motion is a cause of direct irradiance variations at ground level and determines significant fluctuations of PV generation. In this work, we investigate on how integrating information on clouds motion extracted from all-sky images into a time series-based forecasting tool for global horizontal irradiance (GHI) to enhance its prediction performance. We consider three different cloud motion algorithms: heuristic motion detection (HMD), particle image velocimetry (PIV), and a persistent model. The HMD method is originally proposed in this paper. It consists in choosing the cloud motion vector leading to the best cloud map prediction considering the most recent sky images. Results show that integrating the information of the predicted cloud coverage in the circumsolar area leads to a decrease of the width of the GHI prediction intervals up to 2% for prediction horizons in the range 1 to 10 minutes.