For abdominal MRI segmentation, it is difficult to extract the rich information due to the lack of annotated MRI scans. To establish a model of abdominal MRI organ segmentation without MRI annotation, researchers have explored unsupervised cross-modality domain adaptation tasks for abdominal organ segmentation in MRI scans. And our main idea is to rephrase the unsupervised domain adaptive segmentation problem as an image generation problem and a segmentation problem by a two-stage framework. In the first stage, existing methods usually use generative networks to reduce domain gap and cannot consider the intra-domain gap of the target domain. To solve this problem, we propose a single-content multi-style generative network to obtain the multi-style of the target domain rather than the average style. In the second stage, we propose a more simplified pseudo-label selection method to use unlabeled MRI scans. Experiments on the FLARE24 challenge Task3 show that our method achieved an average score of 63.41% and 68.08% for the lesion DSC and NSD on the validation dataset, respectively. The average running time and area under the GPU memory-time curve are 10.36 s and 13331 MB, respectively. Our method not only focuses on the intra-domain gap but also greatly saves resources in the training phase. Our code will be available at https://github.com/ZZhangZZheng/FLARE24-TASK3 .