We construct FarmSeg-VL, the first high-quality image–text dataset for farmland segmentation. It covers eight agricultural regions across four seasons in China, offering extensive spatiotemporal coverage and fine-grained annotations. This dataset fills the gap in remote sensing image–text datasets for farmland, alleviates the challenge of spatiotemporal heterogeneity in farmland segmentation, and provides valuable data to support large-scale farmland monitoring and mapping.