In this work, three models are built to produce intra-day probabilistic solar forecasts with lead times ranging from 10 min to 3 h with a granularity of 10 min. The first model makes only use of past ground measurements. The second model upgrades the first one by adding a variability metric obtained also from the past ground measurements. The third model takes as additional input the satellite albedo. A non parametric approach based on the linear quantile regression technique is used to generate the set of quantiles that summarize the predictive distributions of the global solar irradiance at a horizontal plane (GHI). The probabilistic models are evaluated on several sites that experience very different climatic conditions. It is shown that incorporating variability significantly reduces the width of interval predictions. The addition of satellite information further improves the quality of the probabilistic forecasts.