Steatosis is a common liver disease characterized by the accumulation of lipid droplets in cells. Precise and reliable fat droplet identification is essential for automatic steatosis quantification in histological images. We trained a nnU-Net to automatically segment lipid vacuoles in whole-slide images using semi-automatically generated reference annotations. We evaluated the performance of the trained model on two out-of-distribution datasets. The trained model’s average F1 scores (0.801 and 0.804) suggest a high potential of the nnU-Net framework for the automatic segmentation of lipid vacuoles.