Interactive art often struggles to dynamically reflect audience emotions in real time. To address this, we developed Expressive Canvas, a system that integrates emotion detection with real-time image manipulation. This innovation aims to transform audience emotions into dynamic visual art, creating immersive experiences. Our approach employs a pose-to-emotion detection model using various mathematical equations to enhance accuracy. Trained on the FER-2013 dataset, our model achieved a test accuracy of 62.47%, despite significant challenges arising from class imbalances, particularly for the ’Disgust’ emotion. The system performs efficiently, detecting emotions in an average of 8.43 s and manipulating images in just 0.02 s. Field tests of Expressive Canvas across various events demonstrated its consistent performance and reliability. Despite existing challenges, the system successfully transformed audience emotions into visual expressions, enhancing the connection between art and viewers. This study highlights the potential for real-time emotional expression in interactive art. Future work will focus on mitigating class imbalances and optimizing the neural network to improve accuracy and expressiveness. Key terms include emotion detection, pose-to-emotion detection, real-time image manipulation, interactive art, and neural networks.