"Fencing Hallucination" is a multi-screen interactive installation that merges real-time human-AI interaction in a virtual fencing game with the co-creation of chronophotographs with image-generative AI, documenting audience movements and AI fencer responses. It effectively navigates the challenges of balancing interactivity, modality diversity, and computational constraints within creative AI tools.
The system uses audience pose data as input for a Multilayer Perceptron (MLP) to generate the virtual AI Fencer's pose data. It also leverages this audience pose data to synthesize chronophotographs. This process involves representing pose data as stick figures, using a Dreambooth-fine-tuned Stable Diffusion model for ControlNet-assisted image-to-image translations to transform these stick figures into a series of realistic fencing images, and finally, combining these images with an additive effect to create the final result. This multi-step approach ensures the preservation of overall motion patterns and delicate details when synthesizing a chronophotograph.
"Fencing Hallucination" explores diversity from various angles. Fencing is a unique sport where participants wear pure white attire, rendering their identities anonymous and transcending factors like gender, age, and ethnicity. This feature offers an equitable platform for male and female competitors and minimizes the impact of age compared to other sports.
Moreover, the project contributes to diversifying AI training datasets, addressing the issue of dataset monopolization by organizations in power. It makes fencing datasets more accessible and empowers individuals to create customized datasets.
Additionally, "Fencing Hallucination" fosters a culture of diverse AI creations. It draws inspiration from the tradition of Chronophotography in Photography and uses AI-based image generation to revive the visual aesthetics and photographic experiences of the past. By juxtaposing AI image synthesis and chronophotography, it invites the audience to reflect on the evolution of technology, echoing the innovative approaches of artists like Tom White and Anna Ridler, who blend AI tools with traditional techniques and cultural elements in their artwork.
This project, through its technical innovations and cultural exploration, contributes to the advancement of AI in the context of art and technology. It emphasizes the importance of diversity, inclusivity, and accessibility in this domain, furthering the dialogue on the intersection of AI and artistic expression.