Entity alignment (EA) aims to identify equivalent entities across distinct knowledge graphs (KGs). While existing methods leverage either temporal, structural, or visual information independently, they overlook the synergistic integration of multimodal data. This limitation leads to the absence of dedicated multimodal temporal knowledge graph (MTKG) benchmarks. Current MTKGs exhibit imperfections and are continuously updated; thus, we aim to address these limitations through alignment techniques. Moreover, most state-of-the-art approaches rely on simplistic fusion strategies—such as direct concatenation, averaging, or element-wise addition—to combine multimodal features, resulting in the incomplete utilization of information, leading to the alignment effect falling short of expectations. To address these gaps, we introduce ICEWS+, a novel multimodal temporal KG entity alignment dataset extending the ICEWS benchmark, and propose DynEA: a contrastive learning-based fusion framework enhanced by adaptive attention mechanisms. Empirical evaluations on ICEWS+ demonstrate that DynEA achieves a Hits@1 score of 95.41%, significantly outperforming existing methods.