I have used very high-resolution (30 cm) imagery from the WorldView-3 satellite to track whales in regions where traditional surveys are challenging, including breeding Southern right whales (Eubalaena australis) off the coast of South Africa (images on left) and Beluga whales (Delphinapterus leucas) in the Canadian Arctic (image on right). By training deep learning models (YOLOv12), we can automatically identify individual whales in satellite images, even in remote ice-covered waters. Because training data are limited, we have generated synthetic examples using diffusion models to feed into training so as to improve detection accuracy. This approach has the potential to support population assessments and track seasonal movements with the goal of helping reduce risks from threats such as ship strikes and industrial activity. An academic paper reporting our findings on improving detection performance using diffusion models is forthcoming.