Time series medical imaging can be used to track changes in lesions or body structures over a period of time for diagnosis and treatment planning. In practical scenarios, acquiring time-series medical images is challenging. Generation of time-series image provides an effective solution to this problem. However, existing image generation methods face challenges in generating 3D time-series medical images, primarily due to the lack of effective models that explicitly constrain spatiotemporal relationships. In this paper, we proposed a deformation based spatial-temporal medical image generation method that can generate 3D time-series images between two real 3D images. Specifically, we introduced a two-stage framework. The first stage involves generating spatially consistent 3D image sequences using a deformation-based diffusion model. In the second stage, we enhanced the temporal consistency of these sequences by training a temporal consistency tuning block. Additionally, we designed an optional image quality enhancement module to refine details in images. We validated the proposed method on three public medical datasets of different modalities to generate temporal volumes, demonstrating the efficacy of our method. Compared to existing methods, our method exhibits superior performance in spatiotemporal consistency and image details. Furthermore, experiments demonstrate that using the generated time-series images improves performance in downstream prediction tasks.