Accurate detection of clouds and shadows present in optical imagery is important in remote sensing for ensuring data quality and reliability. This study introduces a deep learning model capable of generating precise cloud and shadows masks for subsequent filtering. Unlike other works in literature, this model operates efficiently across diverse temporalities, sensors, and spatial resolutions, without the need for any relative or absolute transformation of the original data. This versatility, to date unreported in the literature, marks a significant advancement in the field. The model utilizes data from PlanetScope, Landsat and Sentinel-2 sensors and is based on a simplified convolutional neural network (CNN) architecture, LeNet, which facilitates easy training on standard computers with minimal time requirements. Despite its simplicity, the model demonstrates robustness, achieving accuracy metrics over 96% in validation data. These results show the model potential in transforming cloud and shadow detection in remote sensing, combining ease of use with high accuracy.