Fencing Hallucination is a multi-screen interactive installation that enables real-time human-AI interaction in the form of a Fencing game and generates a chronophotograph based on the audience’s movement. It mitigates the conflicts between interactivity, modality variety, and computational limitation in creative AI tools. Fencing Hallucination captures the audience’s pose data as an input to the Multilayer Perceptron(MLP), which generates the virtual AI Fencer’s pose data. It also uses the audience’s pose to synthesize the chronophotograph. The system first represents pose data as stick figures. Then it uses a diffusion model to perform image-to-image translations, converting the stick figures into a series of realistic fencing images. Finally, it combines all images with an additive effect into one image as the result. This multi-step process overcomes the challenge of preserving both the overall motion patterns and fine details when synthesizing a chronophotograph.