The paper introduces a novel generative artificial intelligence (GAI)-assisted human flow detection system (G-HFD) that leverages channel state information (CSI) to estimate the velocity and acceleration of human-induced reflections (HIR). The system employs a unified weighted conditional diffusion model (UW-CDM) to denoise the estimated results, enabling the detection of the number of targets. Additionally, the CSI obtained from a uniform linear array with a wavelength spacing is used to estimate the time of flight (ToF) and direction of arrival (DoA) of the HIR. The UW-CDM addresses the issue of ambiguous DoA spectrum, ensuring accurate DoA estimation. Finally, clustering techniques are used to determine the number of subflows and the subflow size. The evaluation, based on practical downlink communication signals, shows that G-HFD can achieve a subflow size detection accuracy of 91%, validating its effectiveness and highlighting the potential of GAI in wireless sensing.The paper introduces a novel generative artificial intelligence (GAI)-assisted human flow detection system (G-HFD) that leverages channel state information (CSI) to estimate the velocity and acceleration of human-induced reflections (HIR). The system employs a unified weighted conditional diffusion model (UW-CDM) to denoise the estimated results, enabling the detection of the number of targets. Additionally, the CSI obtained from a uniform linear array with a wavelength spacing is used to estimate the time of flight (ToF) and direction of arrival (DoA) of the HIR. The UW-CDM addresses the issue of ambiguous DoA spectrum, ensuring accurate DoA estimation. Finally, clustering techniques are used to determine the number of subflows and the subflow size. The evaluation, based on practical downlink communication signals, shows that G-HFD can achieve a subflow size detection accuracy of 91%, validating its effectiveness and highlighting the potential of GAI in wireless sensing.