In this paper, we introduce sarcasm detection by aiming to increase accuracy and precision to allocate sarcastic content, explain the importance of this model, and then represent the problem faced by sarcasm detection. This research aims to enhance sarcasm detection accuracy using the base of FiLMing Multimodal Sarcasm Detection model, developed by Sundesh Gupta et al. in 2021, which leverages around 24000 image-text pairs and textual attributes processed by RoBERTA (Robustly Optimized BERT approach). By improving the performance and extending its possibilities, I investigated several possible ways of improvements in detection performance.