Collecting 3D point cloud data is cumbersome, so generating high-quality point clouds from existing data can save time and resources while providing more data to support tasks in various fields. In this paper, we propose a neighborhood feature enhancement flow diffusion model for point cloud generation. First, we constructed a multi-scale neighborhood feature aggregation module, which utilizes k-nearest neighbors sampling at different scales to obtain the neighborhood coordinates of each point, thereby aggregating them into coarse global features. Second, we develop a neighborhood attention-based feature enhancement module that uses geometric information in the neighborhood coordinate space to enhance coarse features in the feature space. Then, we used a point-voxel convolutional neural network to reduce redundant features in the enhancement features and output the latent vector of the point cloud. Finally, we transform the latent vectors into data-consistent prior flow features using our designed feature-to-flow data transformation module, seamlessly integrating them into the denoising diffusion model for accurate generation from noisy point clouds. This prior flow approach improves the consistency and coherence of point cloud density distribution. Extensive experiments on the ShapeNet dataset validate the effectiveness of the model in generating 3D point clouds.