This paper presents a novel multimodal summarization and image retrieval framework, termed Context-Aware CLIP with Hybrid Image-Text Fusion (CA-CLIP-HIT), for generating semantically coherent multimodal abstractive summaries and retrieving contextually relevant images. Existing multimodal summarization approaches often exhibit limitations in contextual understanding, semantic alignment between textual and visual modalities, and retrieval accuracy due to shallow feature representations and insufficient cross-modal interaction. To address these challenges, the proposed framework integrates textual and visual information through a Hybrid Image-Text (HIT) fusion mechanism, which generates context-aware multimodal representations capable of preserving semantic dependencies and contextual relationships across modalities. The extracted multimodal features are fused within the HIT layer to construct a comprehensive contextual representation, which is subsequently processed using a sequence-to-sequence encoder-decoder architecture based on Long Short-Term Memory (LSTM) networks for multimodal abstractive summary generation. To strengthen semantic correspondence between generated summaries (SUMMA) and candidate images, the framework employs the CLIP text encoder for summary representation and the CLIP image encoder for visual representation within a shared semantic embedding space. Cosine similarity is utilized to identify and retrieve the image that exhibits the highest semantic relevance to the generated multimodal summary. In addition, a Caption-Assisted Post-Processing Alignment (CAPPA) mechanism is introduced to address ambiguity during image retrieval. In scenarios where multiple candidate images obtain similar similarity scores, the CAPPA module performs caption-level semantic analysis to determine the image that demonstrates the highest contextual alignment with the generated summary. This refinement mechanism enhances retrieval precision, semantic consistency, and robustness in visually ambiguous retrieval scenarios. The proposed framework is evaluated on the MSMO dataset using ROUGE-1, ROUGE-2, ROUGE-L, BLEU, and Image Precision (IP) metrics. Experimental results demonstrate that the proposed CA-CLIP-HIT framework outperforms existing state-of-the-art methods. Furthermore, statistical significance analysis using a paired Student’s t-test confirms that the observed performance improvements are statistically significant. Overall, the proposed CA-CLIP-HIT framework provides an effective and scalable solution for multimodal abstractive summarization and context-aware image retrieval.