The rapid expansion of multimedia content has made video summarization essential for proper content management. The traditional extractive approach which depends on selecting keyframes or segments proves inadequate for understanding video content beyond basic context and narrative structures. Deep learning transformer models including BERT, GPT and T5 have revolutionized abstractive summarization through their self-attention mechanisms. This research investigates the combination of LangChain for managing Large Language Models (LLMs) and Lamini for enhancing summarization quality. The proposed method encounters multiple obstacles because speech-to-text accuracy suffers from background noise and accent variability and video transcript length restrictions and operating expenses from executing big LLMs. The process becomes more complex because maintaining coherence between segmented transcripts and reducing potential biases in AI-generated summaries proves difficult. The method shows certain restrictions because it depends on pre-trained models and lacks visual and emotional comprehension and provides restricted multilingual functionality. The performance evaluation uses metric-based analysis between Lamini and LLaMA models where LLaMA produces better output quality at 85% than Lamini’s 75% and achieves higher accuracy at 90% than Lamini’s 85% but Lamini uses less computation at 100% than LLaMA’s 80%. The results show the need for future work on multimodal integration, domain-specific fine-tuning, and real-time summarization capabilities to address the trade-offs between quality and efficiency.