Solar nowcasting over the 0–4-h horizon is essential to the intra-day scheduling of power systems with high solar penetration. It is, now, widely acknowledged that methods leveraging the geostationary weather satellite data are the most promising options, for such data is able to offer spatio-temporal information that is absolutely vital to capturing the variability of irradiance and thus solar power. Satellite-based irradiance nowcasting takes three general steps: (1) irradiance retrieval from the top-of-the-atmosphere reflectance images, (2) time-forward advection of the irradiance field, and (3) post-processing the forecasts. Since nowcasting applications demand some irradiance retrieval techniques that are computationally light, semi-empirical algorithms such as Heliosat-2 are often favored. On the other hand, optical flow, contrasting other means of acquiring the cloud motion vectors, is commonly regarded as more competitive. On this point, this study presents a concise technical review of several fundamental optical flow algorithms and demonstrates them, with data from Fengyun-4A, which is China’s latest-generation geostationary weather satellite that has hitherto been somewhat under-utilized for solar energy meteorology. Using high-quality measurements from a research-grade radiometric station as verifications, it is found that the 0–4-h nowcasting yields 19.5%–26.7% and 42.4%–53.2% nRMSEs for global horizontal irradiance and beam normal irradiance, respectively, outperforming previous results obtained with Fengyun-4A. Furthermore, ensemble optical flow, which acts as a form of post-processing, is emphasized, as combining the outcomes generated by several peers can shrink the forecast errors of the two irradiance components by 2.2% and 3.0%.