Fairness in recommendation systems has attracted growing attention. In real-world scenarios, user behavior often correlates with sensitive attributes (e.g., gender, age), leading models to amplify biases and cause unfairness. Diffusion models (DMs), known for modeling uncertainty and diversity, align well with the biased nature of real-world recommendations. In this paper, we propose FairGENRec, a Fair GENerative sequential Recommendation model based on DM. During training, we inject noise into user-item distributions under the guidance of a sensitive feature recognition model and design a sequential denoising process for item reconstruction. To mitigate bias, multi-interest representations that remove sensitive feature correlations are incorporated into generation. During inference, FairGENRec reconstructs target items through reverse diffusion from user interaction histories. Extensive experiments on three datasets demonstrate that FairGENRec simultaneously improves both recommendation accuracy and fairness, with case analyses further visualizing fairness enhancements.