In recent years, action transfer for image animation has gained significant attention across various fields. Its core objective is to synthesize realistic video sequences by transferring actions from a driving image to a source image, while meticulously preserving the source’s textural integrity. However, a primary challenge in this domain lies in simultaneously maintaining the source image’s textural and accurately reflecting the driving image’s pose information. To address this, this paper proposes a novel unsupervised image generation framework, namely Dual-Branch Fusion (DBF) network, which comprises a flow-alignment branch and an attention-alignment branch. The flow branch utilizes convolutional warping to preserve local texture consistency, while the attention branch leverages global context modeling to enhance structural alignment. To effectively integrate the complementary strengths of both branches, a cross-fusion module is further introduced to fuse their outputs in a content-aware manner. This design enables DBF to generate high-quality and coherent image sequences that simultaneously retain source appearance and follow target motion. Extensive experiments demonstrate that the proposed method can generate images with relatively realistic structures and textures.