The variability of solar power generation introduces significant challenges to power system operation. Accurate irradiance forecasts will help to overcome these challenges. Therefore, onsite-irradiance forecasting technique was proposed by utilizing a wide-angle camera. The proposed model predicts solar power by several minutes in advance via extracting the cloud motion vectors from sky images. Individual cloud movement was tracked by cross-correlation and motion vectors were created for each cloud. A separate model for cloud deformation and a dynamic model for cloud movement were constructed. For each cloud, cloud deformation over time was identified, and it was embedded into the cloud motion-based forecasts. To improve forecasting results, the cloud motion-based forecasts and time series forecasts were merged. The results were obtained for 1 min, 2min, and 5min horizons with one-minute granularity. The MBE percentages of 11%, 10% and 12% were achieved. The effects of seasonal variations on image-based forecasting models were discussed at the end.