A hybrid forecasting methodology to predict one-minute averaged solar irradiance one to ten minutes in advance is presented and evaluated. The methodology combines the use of all-sky images and irradiance measurements which are both processed in real time to produce the forecast. Pre-existing image processing techniques are locally adapted to estimate the mean motion of clouds, which is used to predict the future sun disk cover by clouds. Then, the predicted cloud information is converted into a solar irradiance estimate using the proposed model which uses real time measurements to extract its parameters for prediction. The validation of the method is done with a sample of 5238 forecasting time points, spread over a six-month period. The forecast uncertainty is assessed separately for clear, cloudy and partly cloudy days, revealing important characteristics of the model's performance under the different conditions. Under partly cloudy and highly variable conditions, positive forecasting skills with respect to regular persistence are achieved above forecasting horizons of two minutes, with a peak performance of 11.4% for forecasting horizons of six and ten minutes. The proposed model also outperforms a smart persistence model for all time horizons under these sky conditions. The model's ramp detection index (RDI, as defined in Chu et al. (2015)) is also evaluated for high and moderate ramps, achieving RDI indexes between 55 and 62% and between 43 and 49% for high and moderate ramps, respectively. These results show that in challenging highly variable solar irradiance conditions the proposed model is suitable for the very short term solar resource forecasting. •All sky images are used to estimate the motion of clouds relative to the sun disk.•Cloud motion and online irradiance data are combined to forecast solar irradiance.•The proposed method is used for very short term forecasting, 1–10 min in advance.•Positive forecasting skills as compared to a persistence model are achieved.